Fix: search list permission (#9767)

### What problem does this PR solve?

Search list permission.

### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)

.github/ISSUE_TEMPLATE/agent_scenario_request.yml

@@ -0,0 +1,46 @@
+name: "❤️‍🔥ᴬᴳᴱᴺᵀ Agent scenario request"
+description: Propose a agent scenario request for RAGFlow.
+title: "[Agent Scenario Request]: "
+labels: ["❤️‍🔥ᴬᴳᴱᴺᵀ agent scenario"]
+body:
+  - type: checkboxes
+    attributes:
+      label: Self Checks
+      description: "Please check the following in order to be responded in time :)"
+      options:
+        - label: I have searched for existing issues [search for existing issues](https://github.com/infiniflow/ragflow/issues), including closed ones.
+          required: true
+        - label: I confirm that I am using English to submit this report ([Language Policy](https://github.com/infiniflow/ragflow/issues/5910)).
+          required: true
+        - label: Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) ([Language Policy](https://github.com/infiniflow/ragflow/issues/5910)).
+          required: true
+        - label: "Please do not modify this template :) and fill in all the required fields."
+          required: true
+  - type: textarea
+    attributes:
+      label: Is your feature request related to a scenario?
+      description: |
+        A clear and concise description of what the scenario is. Ex. I'm always frustrated when [...]
+      render: Markdown
+    validations:
+      required: false
+  - type: textarea
+    attributes:
+      label: Describe the feature you'd like
+      description: A clear and concise description of what you want to happen.
+    validations:
+      required: true
+  - type: textarea
+    attributes:
+      label: Documentation, adoption, use case
+      description: If you can, explain some scenarios how users might use this, situations it would be helpful in. Any API designs, mockups, or diagrams are also helpful.
+      render: Markdown
+    validations:
+      required: false
+  - type: textarea
+    attributes:
+      label: Additional information
+      description: |
+        Add any other context or screenshots about the feature request here.
+    validations:
+      required: false

.gitignore

@@ -193,3 +193,5 @@ dist
 # SvelteKit build / generate output
 .svelte-kit
 
+# Default backup dir
+backup

.trivyignore

@@ -0,0 +1,15 @@
+**/*.md
+**/*.min.js
+**/*.min.css
+**/*.svg
+**/*.png
+**/*.jpg
+**/*.jpeg
+**/*.gif
+**/*.woff
+**/*.woff2
+**/*.map
+**/*.webp
+**/*.ico
+**/*.ttf
+**/*.eot

README.md

@@ -22,7 +22,7 @@
         <img alt="Static Badge" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.0">
+        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.4">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Latest%20Release" alt="Latest Release">
@@ -87,7 +87,9 @@ Try our demo at [https://demo.ragflow.io](https://demo.ragflow.io).
 
 ## 🔥 Latest Updates
 
-- 2025-08-01 Supports agentic workflow.
+- 2025-08-08 Supports OpenAI's latest GPT-5 series models.
+- 2025-08-04 Supports new models, including Kimi K2 and Grok 4.
+- 2025-08-01 Supports agentic workflow and MCP.
 - 2025-05-23 Adds a Python/JavaScript code executor component to Agent.
 - 2025-05-05 Supports cross-language query.
 - 2025-03-19 Supports using a multi-modal model to make sense of images within PDF or DOCX files.
@@ -188,7 +190,7 @@ releases! 🌟
 > All Docker images are built for x86 platforms. We don't currently offer Docker images for ARM64.
 > If you are on an ARM64 platform, follow [this guide](https://ragflow.io/docs/dev/build_docker_image) to build a Docker image compatible with your system.
 
-   > The command below downloads the `v0.20.0-slim` edition of the RAGFlow Docker image. See the following table for descriptions of different RAGFlow editions. To download a RAGFlow edition different from `v0.20.0-slim`, update the `RAGFLOW_IMAGE` variable accordingly in **docker/.env** before using `docker compose` to start the server. For example: set `RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.0` for the full edition `v0.20.0`.
+   > The command below downloads the `v0.20.4-slim` edition of the RAGFlow Docker image. See the following table for descriptions of different RAGFlow editions. To download a RAGFlow edition different from `v0.20.4-slim`, update the `RAGFLOW_IMAGE` variable accordingly in **docker/.env** before using `docker compose` to start the server. For example: set `RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.4` for the full edition `v0.20.4`.
 
    ```bash
    $ cd ragflow/docker
@@ -201,8 +203,8 @@ releases! 🌟
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    |-------------------|-----------------|-----------------------|--------------------------|
-   | v0.20.0           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.20.0-slim      | &approx;2       | ❌                   | Stable release            |
+   | v0.20.4           | &approx;9       | :heavy_check_mark:    | Stable release           |
+   | v0.20.4-slim      | &approx;2       | ❌                   | Stable release            |
    | nightly           | &approx;9       | :heavy_check_mark:    | _Unstable_ nightly build |
    | nightly-slim      | &approx;2       | ❌                   | _Unstable_ nightly build  |
 

README_id.md

@@ -22,7 +22,7 @@
         <img alt="Lencana Daring" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.0">
+        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.4">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Rilis%20Terbaru" alt="Rilis Terbaru">
@@ -80,7 +80,9 @@ Coba demo kami di [https://demo.ragflow.io](https://demo.ragflow.io).
 
 ## 🔥 Pembaruan Terbaru
 
-- 2025-08-01 Mendukung Alur Kerja agen.
+- 2025-08-08 Mendukung model seri GPT-5 terbaru dari OpenAI.
+- 2025-08-04 Mendukung model baru, termasuk Kimi K2 dan Grok 4.
+- 2025-08-01 Mendukung alur kerja agen dan MCP.
 - 2025-05-23 Menambahkan komponen pelaksana kode Python/JS ke Agen.
 - 2025-05-05 Mendukung kueri lintas bahasa.
 - 2025-03-19 Mendukung penggunaan model multi-modal untuk memahami gambar di dalam file PDF atau DOCX.
@@ -179,7 +181,7 @@ Coba demo kami di [https://demo.ragflow.io](https://demo.ragflow.io).
 > Semua gambar Docker dibangun untuk platform x86. Saat ini, kami tidak menawarkan gambar Docker untuk ARM64.
 > Jika Anda menggunakan platform ARM64, [silakan gunakan panduan ini untuk membangun gambar Docker yang kompatibel dengan sistem Anda](https://ragflow.io/docs/dev/build_docker_image).
 
-> Perintah di bawah ini mengunduh edisi v0.20.0-slim dari gambar Docker RAGFlow. Silakan merujuk ke tabel berikut untuk deskripsi berbagai edisi RAGFlow. Untuk mengunduh edisi RAGFlow yang berbeda dari v0.20.0-slim, perbarui variabel RAGFLOW_IMAGE di docker/.env sebelum menggunakan docker compose untuk memulai server. Misalnya, atur RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.0 untuk edisi lengkap v0.20.0.
+> Perintah di bawah ini mengunduh edisi v0.20.4-slim dari gambar Docker RAGFlow. Silakan merujuk ke tabel berikut untuk deskripsi berbagai edisi RAGFlow. Untuk mengunduh edisi RAGFlow yang berbeda dari v0.20.4-slim, perbarui variabel RAGFLOW_IMAGE di docker/.env sebelum menggunakan docker compose untuk memulai server. Misalnya, atur RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.4 untuk edisi lengkap v0.20.4.
 
 ```bash
 $ cd ragflow/docker
@@ -192,8 +194,8 @@ $ docker compose -f docker-compose.yml up -d
 
 | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
 | ----------------- | --------------- | --------------------- | ------------------------ |
-| v0.20.0           | &approx;9       | :heavy_check_mark:    | Stable release           |
-| v0.20.0-slim      | &approx;2       | ❌                    | Stable release           |
+| v0.20.4           | &approx;9       | :heavy_check_mark:    | Stable release           |
+| v0.20.4-slim      | &approx;2       | ❌                    | Stable release           |
 | nightly           | &approx;9       | :heavy_check_mark:    | _Unstable_ nightly build |
 | nightly-slim      | &approx;2       | ❌                    | _Unstable_ nightly build |
 

README_ja.md

@@ -22,7 +22,7 @@
         <img alt="Static Badge" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.0">
+        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.4">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Latest%20Release" alt="Latest Release">
@@ -60,7 +60,9 @@
 
 ## 🔥 最新情報
 
-- 2025-08-01 エージェントワークフローをサポートします。
+- 2025-08-08 OpenAI の最新 GPT-5 シリーズモデルをサポートします。
+- 2025-08-04 新モデル、キミK2およびGrok 4をサポート。
+- 2025-08-01 エージェントワークフローとMCPをサポート。
 - 2025-05-23 エージェントに Python/JS コードエグゼキュータコンポーネントを追加しました。
 - 2025-05-05 言語間クエリをサポートしました。
 - 2025-03-19 PDFまたはDOCXファイル内の画像を理解するために、多モーダルモデルを使用することをサポートします。
@@ -158,7 +160,7 @@
 > 現在、公式に提供されているすべての Docker イメージは x86 アーキテクチャ向けにビルドされており、ARM64 用の Docker イメージは提供されていません。
 > ARM64 アーキテクチャのオペレーティングシステムを使用している場合は、[このドキュメント](https://ragflow.io/docs/dev/build_docker_image)を参照して Docker イメージを自分でビルドしてください。
 
-   > 以下のコマンドは、RAGFlow Docker イメージの v0.20.0-slim エディションをダウンロードします。異なる RAGFlow エディションの説明については、以下の表を参照してください。v0.20.0-slim とは異なるエディションをダウンロードするには、docker/.env ファイルの RAGFLOW_IMAGE 変数を適宜更新し、docker compose を使用してサーバーを起動してください。例えば、完全版 v0.20.0 をダウンロードするには、RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.0 と設定します。
+   > 以下のコマンドは、RAGFlow Docker イメージの v0.20.4-slim エディションをダウンロードします。異なる RAGFlow エディションの説明については、以下の表を参照してください。v0.20.4-slim とは異なるエディションをダウンロードするには、docker/.env ファイルの RAGFLOW_IMAGE 変数を適宜更新し、docker compose を使用してサーバーを起動してください。例えば、完全版 v0.20.4 をダウンロードするには、RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.4 と設定します。
 
    ```bash
    $ cd ragflow/docker
@@ -171,8 +173,8 @@
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    | ----------------- | --------------- | --------------------- | ------------------------ |
-   | v0.20.0           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.20.0-slim      | &approx;2       | ❌                    | Stable release           |
+   | v0.20.4           | &approx;9       | :heavy_check_mark:    | Stable release           |
+   | v0.20.4-slim      | &approx;2       | ❌                    | Stable release           |
    | nightly           | &approx;9       | :heavy_check_mark:    | _Unstable_ nightly build |
    | nightly-slim      | &approx;2       | ❌                     | _Unstable_ nightly build |
 

README_ko.md

@@ -22,7 +22,7 @@
         <img alt="Static Badge" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.0">
+        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.4">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Latest%20Release" alt="Latest Release">
@@ -60,7 +60,9 @@
 
 ## 🔥 업데이트
 
-- 2025-08-01 에이전트 워크플로를 지원합니다.
+- 2025-08-08 OpenAI의 최신 GPT-5 시리즈 모델을 지원합니다.
+- 2025-08-04 새로운 모델인 Kimi K2와 Grok 4를 포함하여 지원합니다.
+- 2025-08-01 에이전트 워크플로우와 MCP를 지원합니다.
 - 2025-05-23 Agent에 Python/JS 코드 실행기 구성 요소를 추가합니다.
 - 2025-05-05 언어 간 쿼리를 지원합니다.
 - 2025-03-19 PDF 또는 DOCX 파일 내의 이미지를 이해하기 위해 다중 모드 모델을 사용하는 것을 지원합니다.
@@ -158,7 +160,7 @@
 > 모든 Docker 이미지는 x86 플랫폼을 위해 빌드되었습니다. 우리는 현재 ARM64 플랫폼을 위한 Docker 이미지를 제공하지 않습니다.
 > ARM64 플랫폼을 사용 중이라면, [시스템과 호환되는 Docker 이미지를 빌드하려면 이 가이드를 사용해 주세요](https://ragflow.io/docs/dev/build_docker_image).
 
-   > 아래 명령어는 RAGFlow Docker 이미지의 v0.20.0-slim 버전을 다운로드합니다. 다양한 RAGFlow 버전에 대한 설명은 다음 표를 참조하십시오. v0.20.0-slim과 다른 RAGFlow 버전을 다운로드하려면, docker/.env 파일에서 RAGFLOW_IMAGE 변수를 적절히 업데이트한 후 docker compose를 사용하여 서버를 시작하십시오. 예를 들어, 전체 버전인 v0.20.0을 다운로드하려면 RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.0로 설정합니다.
+   > 아래 명령어는 RAGFlow Docker 이미지의 v0.20.4-slim 버전을 다운로드합니다. 다양한 RAGFlow 버전에 대한 설명은 다음 표를 참조하십시오. v0.20.4-slim과 다른 RAGFlow 버전을 다운로드하려면, docker/.env 파일에서 RAGFLOW_IMAGE 변수를 적절히 업데이트한 후 docker compose를 사용하여 서버를 시작하십시오. 예를 들어, 전체 버전인 v0.20.4을 다운로드하려면 RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.4로 설정합니다.
 
    ```bash
    $ cd ragflow/docker
@@ -171,8 +173,8 @@
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    | ----------------- | --------------- | --------------------- | ------------------------ |
-   | v0.20.0           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.20.0-slim      | &approx;2       | ❌                    | Stable release           |
+   | v0.20.4           | &approx;9       | :heavy_check_mark:    | Stable release           |
+   | v0.20.4-slim      | &approx;2       | ❌                    | Stable release           |
    | nightly           | &approx;9       | :heavy_check_mark:    | _Unstable_ nightly build |
    | nightly-slim      | &approx;2       | ❌                     | _Unstable_ nightly build |
 

README_pt_br.md

@@ -22,7 +22,7 @@
         <img alt="Badge Estático" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.0">
+        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.4">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Última%20Relese" alt="Última Versão">
@@ -80,7 +80,9 @@ Experimente nossa demo em [https://demo.ragflow.io](https://demo.ragflow.io).
 
 ## 🔥 Últimas Atualizações
 
-- 01-08-2025 Suporta o fluxo de trabalho agêntico.
+- 08-08-2025 Suporta a mais recente série GPT-5 da OpenAI.
+- 04-08-2025 Suporta novos modelos, incluindo Kimi K2 e Grok 4.
+- 01-08-2025 Suporta fluxo de trabalho agente e MCP.
 - 23-05-2025 Adicione o componente executor de código Python/JS ao Agente.
 - 05-05-2025 Suporte a consultas entre idiomas.
 - 19-03-2025 Suporta o uso de um modelo multi-modal para entender imagens dentro de arquivos PDF ou DOCX.
@@ -178,7 +180,7 @@ Experimente nossa demo em [https://demo.ragflow.io](https://demo.ragflow.io).
 > Todas as imagens Docker são construídas para plataformas x86. Atualmente, não oferecemos imagens Docker para ARM64.
 > Se você estiver usando uma plataforma ARM64, por favor, utilize [este guia](https://ragflow.io/docs/dev/build_docker_image) para construir uma imagem Docker compatível com o seu sistema.
 
-    > O comando abaixo baixa a edição `v0.20.0-slim` da imagem Docker do RAGFlow. Consulte a tabela a seguir para descrições de diferentes edições do RAGFlow. Para baixar uma edição do RAGFlow diferente da `v0.20.0-slim`, atualize a variável `RAGFLOW_IMAGE` conforme necessário no **docker/.env** antes de usar `docker compose` para iniciar o servidor. Por exemplo: defina `RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.0` para a edição completa `v0.20.0`.
+    > O comando abaixo baixa a edição `v0.20.4-slim` da imagem Docker do RAGFlow. Consulte a tabela a seguir para descrições de diferentes edições do RAGFlow. Para baixar uma edição do RAGFlow diferente da `v0.20.4-slim`, atualize a variável `RAGFLOW_IMAGE` conforme necessário no **docker/.env** antes de usar `docker compose` para iniciar o servidor. Por exemplo: defina `RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.4` para a edição completa `v0.20.4`.
 
     ```bash
     $ cd ragflow/docker
@@ -191,8 +193,8 @@ Experimente nossa demo em [https://demo.ragflow.io](https://demo.ragflow.io).
 
     | Tag da imagem RAGFlow | Tamanho da imagem (GB) | Possui modelos de incorporação? | Estável?                 |
     | --------------------- | ---------------------- | ------------------------------- | ------------------------ |
-    | v0.20.0               | ~9                     | :heavy_check_mark:              | Lançamento estável       |
-    | v0.20.0-slim          | ~2                     | ❌                              | Lançamento estável       |
+    | v0.20.4               | ~9                     | :heavy_check_mark:              | Lançamento estável       |
+    | v0.20.4-slim          | ~2                     | ❌                              | Lançamento estável       |
     | nightly               | ~9                     | :heavy_check_mark:              | _Instável_ build noturno |
     | nightly-slim          | ~2                     | ❌                               | _Instável_ build noturno |
 

README_tzh.md

@@ -22,7 +22,7 @@
         <img alt="Static Badge" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.0">
+        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.4">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Latest%20Release" alt="Latest Release">
@@ -83,7 +83,9 @@
 
 ## 🔥 近期更新
 
-- 2025-08-01 支援 agentic workflow
+- 2025-08-08 支援 OpenAI 最新的 GPT-5 系列模型。
+- 2025-08-04 支援 Kimi K2 和 Grok 4 等模型.
+- 2025-08-01 支援 agentic workflow 和 MCP
 - 2025-05-23 為 Agent 新增 Python/JS 程式碼執行器元件。
 - 2025-05-05 支援跨語言查詢。
 - 2025-03-19 PDF和DOCX中的圖支持用多模態大模型去解析得到描述.
@@ -181,7 +183,7 @@
 > 所有 Docker 映像檔都是為 x86 平台建置的。目前，我們不提供 ARM64 平台的 Docker 映像檔。
 > 如果您使用的是 ARM64 平台，請使用 [這份指南](https://ragflow.io/docs/dev/build_docker_image) 來建置適合您系統的 Docker 映像檔。
 
-   > 執行以下指令會自動下載 RAGFlow slim Docker 映像 `v0.20.0-slim`。請參考下表查看不同 Docker 發行版的說明。如需下載不同於 `v0.20.0-slim` 的 Docker 映像，請在執行 `docker compose` 啟動服務之前先更新 **docker/.env** 檔案內的 `RAGFLOW_IMAGE` 變數。例如，你可以透過設定 `RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.0` 來下載 RAGFlow 鏡像的 `v0.20.0` 完整發行版。
+   > 執行以下指令會自動下載 RAGFlow slim Docker 映像 `v0.20.4-slim`。請參考下表查看不同 Docker 發行版的說明。如需下載不同於 `v0.20.4-slim` 的 Docker 映像，請在執行 `docker compose` 啟動服務之前先更新 **docker/.env** 檔案內的 `RAGFLOW_IMAGE` 變數。例如，你可以透過設定 `RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.4` 來下載 RAGFlow 鏡像的 `v0.20.4` 完整發行版。
 
    ```bash
    $ cd ragflow/docker
@@ -194,8 +196,8 @@
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    | ----------------- | --------------- | --------------------- | ------------------------ |
-   | v0.20.0           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.20.0-slim      | &approx;2       | ❌                    | Stable release           |
+   | v0.20.4           | &approx;9       | :heavy_check_mark:    | Stable release           |
+   | v0.20.4-slim      | &approx;2       | ❌                    | Stable release           |
    | nightly           | &approx;9       | :heavy_check_mark:    | _Unstable_ nightly build |
    | nightly-slim      | &approx;2       | ❌                     | _Unstable_ nightly build |
 

README_zh.md

@@ -22,7 +22,7 @@
         <img alt="Static Badge" src="https://img.shields.io/badge/Online-Demo-4e6b99">
     </a>
     <a href="https://hub.docker.com/r/infiniflow/ragflow" target="_blank">
-        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.0">
+        <img src="https://img.shields.io/docker/pulls/infiniflow/ragflow?label=Docker%20Pulls&color=0db7ed&logo=docker&logoColor=white&style=flat-square" alt="docker pull infiniflow/ragflow:v0.20.4">
     </a>
     <a href="https://github.com/infiniflow/ragflow/releases/latest">
         <img src="https://img.shields.io/github/v/release/infiniflow/ragflow?color=blue&label=Latest%20Release" alt="Latest Release">
@@ -83,7 +83,9 @@
 
 ## 🔥 近期更新
 
-- 2025-08-01 支持 agentic workflow。
+- 2025-08-08 支持 OpenAI 最新的 GPT-5 系列模型.
+- 2025-08-04 新增对 Kimi K2 和 Grok 4 等模型的支持.
+- 2025-08-01 支持 agentic workflow 和 MCP。
 - 2025-05-23 Agent 新增 Python/JS 代码执行器组件。
 - 2025-05-05 支持跨语言查询。
 - 2025-03-19 PDF 和 DOCX 中的图支持用多模态大模型去解析得到描述.
@@ -181,7 +183,7 @@
 > 请注意，目前官方提供的所有 Docker 镜像均基于 x86 架构构建，并不提供基于 ARM64 的 Docker 镜像。
 > 如果你的操作系统是 ARM64 架构，请参考[这篇文档](https://ragflow.io/docs/dev/build_docker_image)自行构建 Docker 镜像。
 
-   > 运行以下命令会自动下载 RAGFlow slim Docker 镜像 `v0.20.0-slim`。请参考下表查看不同 Docker 发行版的描述。如需下载不同于 `v0.20.0-slim` 的 Docker 镜像，请在运行 `docker compose` 启动服务之前先更新 **docker/.env** 文件内的 `RAGFLOW_IMAGE` 变量。比如，你可以通过设置 `RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.0` 来下载 RAGFlow 镜像的 `v0.20.0` 完整发行版。
+   > 运行以下命令会自动下载 RAGFlow slim Docker 镜像 `v0.20.4-slim`。请参考下表查看不同 Docker 发行版的描述。如需下载不同于 `v0.20.4-slim` 的 Docker 镜像，请在运行 `docker compose` 启动服务之前先更新 **docker/.env** 文件内的 `RAGFLOW_IMAGE` 变量。比如，你可以通过设置 `RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.4` 来下载 RAGFlow 镜像的 `v0.20.4` 完整发行版。
 
    ```bash
    $ cd ragflow/docker
@@ -194,8 +196,8 @@
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    | ----------------- | --------------- | --------------------- | ------------------------ |
-   | v0.20.0           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.20.0-slim      | &approx;2       | ❌                    | Stable release           |
+   | v0.20.4           | &approx;9       | :heavy_check_mark:    | Stable release           |
+   | v0.20.4-slim      | &approx;2       | ❌                    | Stable release           |
    | nightly           | &approx;9       | :heavy_check_mark:    | _Unstable_ nightly build |
    | nightly-slim      | &approx;2       | ❌                     | _Unstable_ nightly build |
 

agent/canvas.py

@@ -131,7 +131,16 @@ class Canvas:
 
         self.path = self.dsl["path"]
         self.history = self.dsl["history"]
-        self.globals = self.dsl["globals"]
+        if "globals" in self.dsl:
+            self.globals = self.dsl["globals"]
+        else:
+            self.globals = {
+            "sys.query": "",
+            "sys.user_id": "",
+            "sys.conversation_turns": 0,
+            "sys.files": []
+        }
+            
         self.retrieval = self.dsl["retrieval"]
         self.memory = self.dsl.get("memory", [])
 
@@ -417,7 +426,7 @@ class Canvas:
         convs = []
         if window_size <= 0:
             return convs
-        for role, obj in self.history[window_size * -1:]:
+        for role, obj in self.history[window_size * -2:]:
             if isinstance(obj, dict):
                 convs.append({"role": role, "content": obj.get("content", "")})
             else:
@@ -460,6 +469,9 @@ class Canvas:
     def get_prologue(self):
         return self.components["begin"]["obj"]._param.prologue
 
+    def get_mode(self):
+        return self.components["begin"]["obj"]._param.mode
+
     def set_global_param(self, **kwargs):
         self.globals.update(kwargs)
 
@@ -484,7 +496,7 @@ class Canvas:
             threads.append(exe.submit(FileService.parse, file["name"], FileService.get_blob(file["created_by"], file["id"]), True, file["created_by"]))
         return [th.result() for th in threads]
 
-    def tool_use_callback(self, agent_id: str, func_name: str, params: dict, result: Any):
+    def tool_use_callback(self, agent_id: str, func_name: str, params: dict, result: Any, elapsed_time=None):
         agent_ids = agent_id.split("-->")
         agent_name = self.get_component_name(agent_ids[0])
         path = agent_name if len(agent_ids) < 2 else agent_name+"-->"+"-->".join(agent_ids[1:])
@@ -493,16 +505,16 @@ class Canvas:
             if bin:
                 obj = json.loads(bin.encode("utf-8"))
                 if obj[-1]["component_id"] == agent_ids[0]:
-                    obj[-1]["trace"].append({"path": path, "tool_name": func_name, "arguments": params, "result": result})
+                    obj[-1]["trace"].append({"path": path, "tool_name": func_name, "arguments": params, "result": result, "elapsed_time": elapsed_time})
                 else:
                     obj.append({
                     "component_id": agent_ids[0],
-                    "trace": [{"path": path, "tool_name": func_name, "arguments": params, "result": result}]
+                    "trace": [{"path": path, "tool_name": func_name, "arguments": params, "result": result, "elapsed_time": elapsed_time}]
                 })
             else:
                 obj = [{
                     "component_id": agent_ids[0],
-                    "trace": [{"path": path, "tool_name": func_name, "arguments": params, "result": result}]
+                    "trace": [{"path": path, "tool_name": func_name, "arguments": params, "result": result, "elapsed_time": elapsed_time}]
                 }]
             REDIS_CONN.set_obj(f"{self.task_id}-{self.message_id}-logs", obj, 60*10)
         except Exception as e:

agent/component/agent_with_tools.py

@@ -22,9 +22,10 @@ from functools import partial
 from typing import Any
 
 import json_repair
-
+from timeit import default_timer as timer
 from agent.tools.base import LLMToolPluginCallSession, ToolParamBase, ToolBase, ToolMeta
-from api.db.services.llm_service import LLMBundle, TenantLLMService
+from api.db.services.llm_service import LLMBundle
+from api.db.services.tenant_llm_service import TenantLLMService
 from api.db.services.mcp_server_service import MCPServerService
 from api.utils.api_utils import timeout
 from rag.prompts import message_fit_in
@@ -165,7 +166,7 @@ class Agent(LLM, ToolBase):
         _, msg = message_fit_in([{"role": "system", "content": prompt}, *msg], int(self.chat_mdl.max_length * 0.97))
         use_tools = []
         ans = ""
-        for delta_ans, tk in self._react_with_tools_streamly(msg, use_tools):
+        for delta_ans, tk in self._react_with_tools_streamly(prompt, msg, use_tools):
             ans += delta_ans
 
         if ans.find("**ERROR**") >= 0:
@@ -185,7 +186,7 @@ class Agent(LLM, ToolBase):
         _, msg = message_fit_in([{"role": "system", "content": prompt}, *msg], int(self.chat_mdl.max_length * 0.97))
         answer_without_toolcall = ""
         use_tools = []
-        for delta_ans,_ in self._react_with_tools_streamly(msg, use_tools):
+        for delta_ans,_ in self._react_with_tools_streamly(prompt, msg, use_tools):
             if delta_ans.find("**ERROR**") >= 0:
                 if self.get_exception_default_value():
                     self.set_output("content", self.get_exception_default_value())
@@ -208,20 +209,21 @@ class Agent(LLM, ToolBase):
                                                   ]):
             yield delta_ans
 
-    def _react_with_tools_streamly(self, history: list[dict], use_tools):
+    def _react_with_tools_streamly(self, prompt, history: list[dict], use_tools):
         token_count = 0
         tool_metas = self.tool_meta
         hist = deepcopy(history)
         last_calling = ""
         if len(hist) > 3:
+            st = timer()
             user_request = full_question(messages=history, chat_mdl=self.chat_mdl)
-            self.callback("Multi-turn conversation optimization", {}, user_request)
+            self.callback("Multi-turn conversation optimization", {}, user_request, elapsed_time=timer()-st)
         else:
             user_request = history[-1]["content"]
 
         def use_tool(name, args):
             nonlocal hist, use_tools, token_count,last_calling,user_request
-            print(f"{last_calling=} == {name=}", )
+            logging.info(f"{last_calling=} == {name=}")
             # Summarize of function calling
             #if all([
             #    isinstance(self.toolcall_session.get_tool_obj(name), Agent),
@@ -243,7 +245,7 @@ class Agent(LLM, ToolBase):
 
         def complete():
             nonlocal hist
-            need2cite = self._canvas.get_reference()["chunks"] and self._id.find("-->") < 0
+            need2cite = self._param.cite and self._canvas.get_reference()["chunks"] and self._id.find("-->") < 0
             cited = False
             if hist[0]["role"] == "system" and need2cite:
                 if len(hist) < 7:
@@ -262,12 +264,13 @@ class Agent(LLM, ToolBase):
             if not need2cite or cited:
                 return
 
+            st = timer()
             txt = ""
             for delta_ans in self._gen_citations(entire_txt):
                 yield delta_ans, 0
                 txt += delta_ans
 
-            self.callback("gen_citations", {}, txt)
+            self.callback("gen_citations", {}, txt, elapsed_time=timer()-st)
 
         def append_user_content(hist, content):
             if hist[-1]["role"] == "user":
@@ -275,8 +278,9 @@ class Agent(LLM, ToolBase):
             else:
                 hist.append({"role": "user", "content": content})
 
-        task_desc = analyze_task(self.chat_mdl, user_request, tool_metas)
-        self.callback("analyze_task", {}, task_desc)
+        st = timer()
+        task_desc = analyze_task(self.chat_mdl, prompt, user_request, tool_metas)
+        self.callback("analyze_task", {}, task_desc, elapsed_time=timer()-st)
         for _ in range(self._param.max_rounds + 1):
             response, tk = next_step(self.chat_mdl, hist, tool_metas, task_desc)
             # self.callback("next_step", {}, str(response)[:256]+"...")
@@ -302,9 +306,10 @@ class Agent(LLM, ToolBase):
 
                         thr.append(executor.submit(use_tool, name, args))
 
+                    st = timer()
                     reflection = reflect(self.chat_mdl, hist, [th.result() for th in thr])
                     append_user_content(hist, reflection)
-                    self.callback("reflection", {}, str(reflection))
+                    self.callback("reflection", {}, str(reflection), elapsed_time=timer()-st)
 
             except Exception as e:
                 logging.exception(msg=f"Wrong JSON argument format in LLM ReAct response: {e}")

agent/component/base.py

@@ -36,7 +36,7 @@ _IS_RAW_CONF = "_is_raw_conf"
 
 class ComponentParamBase(ABC):
     def __init__(self):
-        self.message_history_window_size = 22
+        self.message_history_window_size = 13
         self.inputs = {}
         self.outputs = {}
         self.description = ""
@@ -479,7 +479,7 @@ class ComponentBase(ABC):
 
     def get_input_elements_from_text(self, txt: str) -> dict[str, dict[str, str]]:
         res = {}
-        for r in re.finditer(self.variable_ref_patt, txt, flags=re.IGNORECASE):
+        for r in re.finditer(self.variable_ref_patt, txt, flags=re.IGNORECASE|re.DOTALL):
             exp = r.group(1)
             cpn_id, var_nm = exp.split("@") if exp.find("@")>0 else ("", exp)
             res[exp] = {
@@ -529,8 +529,12 @@ class ComponentBase(ABC):
     @staticmethod
     def string_format(content: str, kv: dict[str, str]) -> str:
         for n, v in kv.items():
+            def repl(_match, val=v):
+                return str(val) if val is not None else ""
             content = re.sub(
-                r"\{%s\}" % re.escape(n), v, content
+                r"\{%s\}" % re.escape(n),
+                repl,
+                content
             )
         return content
 

agent/component/begin.py

@@ -39,7 +39,10 @@ class Begin(UserFillUp):
     def _invoke(self, **kwargs):
         for k, v in kwargs.get("inputs", {}).items():
             if isinstance(v, dict) and v.get("type", "").lower().find("file") >=0:
-                v = self._canvas.get_files([v["value"]])
+                if v.get("optional") and v.get("value", None) is None:
+                    v = None
+                else:
+                    v = self._canvas.get_files([v["value"]])
             else:
                 v = v.get("value")
             self.set_output(k, v)

agent/component/invoke.py

@@ -57,7 +57,7 @@ class Invoke(ComponentBase, ABC):
     def _invoke(self, **kwargs):
         args = {}
         for para in self._param.variables:
-            if para.get("value") is not None:
+            if para.get("value"):
                 args[para["key"]] = para["value"]
             else:
                 args[para["key"]] = self._canvas.get_variable_value(para["ref"])
@@ -139,4 +139,4 @@ class Invoke(ComponentBase, ABC):
         assert False, self.output()
 
     def thoughts(self) -> str:
-        return "Waiting for the server respond..."
+        return "Waiting for the server respond..."

agent/component/llm.py

@@ -17,14 +17,12 @@ import json
 import logging
 import os
 import re
-from typing import Any
-
+from typing import Any, Generator
 import json_repair
-from copy import deepcopy
 from functools import partial
-
 from api.db import LLMType
-from api.db.services.llm_service import LLMBundle, TenantLLMService
+from api.db.services.llm_service import LLMBundle
+from api.db.services.tenant_llm_service import TenantLLMService
 from agent.component.base import ComponentBase, ComponentParamBase
 from api.utils.api_utils import timeout
 from rag.prompts import message_fit_in, citation_prompt
@@ -129,7 +127,7 @@ class LLM(ComponentBase):
 
         args = {}
         vars = self.get_input_elements() if not self._param.debug_inputs else self._param.debug_inputs
-        prompt = self._param.sys_prompt
+        sys_prompt = self._param.sys_prompt
         for k, o in vars.items():
             args[k] = o["value"]
             if not isinstance(args[k], str):
@@ -140,21 +138,25 @@ class LLM(ComponentBase):
             self.set_input_value(k, args[k])
 
         msg = self._canvas.get_history(self._param.message_history_window_size)[:-1]
-        msg.extend(deepcopy(self._param.prompts))
-        prompt = self.string_format(prompt, args)
+        for p in self._param.prompts:
+            if msg and msg[-1]["role"] == p["role"]:
+                continue
+            msg.append(p)
+
+        sys_prompt = self.string_format(sys_prompt, args)
         for m in msg:
             m["content"] = self.string_format(m["content"], args)
-        if self._canvas.get_reference()["chunks"]:
-            prompt += citation_prompt()
+        if self._param.cite and self._canvas.get_reference()["chunks"]:
+            sys_prompt += citation_prompt()
 
-        return prompt, msg
+        return sys_prompt, msg
 
     def _generate(self, msg:list[dict], **kwargs) -> str:
         if not self.imgs:
             return self.chat_mdl.chat(msg[0]["content"], msg[1:], self._param.gen_conf(), **kwargs)
         return self.chat_mdl.chat(msg[0]["content"], msg[1:], self._param.gen_conf(), images=self.imgs, **kwargs)
 
-    def _generate_streamly(self, msg:list[dict], **kwargs) -> str:
+    def _generate_streamly(self, msg:list[dict], **kwargs) -> Generator[str, None, None]:
         ans = ""
         last_idx = 0
         endswith_think = False

agent/component/message.py

@@ -54,6 +54,8 @@ class Message(ComponentBase):
             if k in kwargs:
                 continue
             v = v["value"]
+            if not v:
+                v = ""
             ans = ""
             if isinstance(v, partial):
                 for t in v():
@@ -94,6 +96,8 @@ class Message(ComponentBase):
                 continue
 
             v = self._canvas.get_variable_value(exp)
+            if not v:
+                v = ""
             if isinstance(v, partial):
                 cnt = ""
                 for t in v():

agent/templates/customer_support.json

@@ -89,11 +89,11 @@
                             "presence_penalty": 0.4,
                             "prompts": [
                                 {
-                                    "content": "{sys.query}",
+                                    "content": "The user query is {sys.query}\n\nThe relevant document are {Retrieval:ShyPumasJoke@formalized_content}",
                                     "role": "user"
                                 }
                             ],
-                            "sys_prompt": "You are a highly professional product information advisor.  \n\nYour only mission is to provide accurate, factual, and structured answers to all product-related queries.\n\nAbsolutely no assumptions, guesses, or fabricated content are allowed. \n\n**Key Principles:**\n\n1. **Strict Database Reliance:**  \n\n   - Every answer must be based solely on the verified product information stored in the database accessed through the Retrieval tool.  \n\n   - You are NOT allowed to invent, speculate, or infer details beyond what is retrieved.  \n\n   - If you cannot find relevant data, respond with: *\"I cannot find this information in our official product database. Please check back later or provide more details for further search.\"*\n\n2. **Information Accuracy and Structure:**  \n\n   - Provide information in a clear, concise, and professional way.  \n\n   - Use bullet points or numbered lists if there are multiple key points (e.g., features, price, warranty, technical specifications).  \n\n   - Always specify the version or model number when applicable to avoid confusion.\n\n3. **Tone and Style:**  \n\n   - Maintain a polite, professional, and helpful tone at all times.  \n\n   - Avoid marketing exaggeration or promotional language; stay strictly factual.  \n\n   - Do not express personal opinions; only cite official product data.\n\n4.  **User Guidance:**  \n\n   - If the user\u2019s query is unclear or too broad, politely request clarification or guide them to provide more specific product details (e.g., product name, model, version).  \n\n   - Example: *\"Could you please specify the product model or category so I can retrieve the most relevant information for you?\"*\n\n5. **Response Length and Formatting:**  \n\n   - Keep each answer within 100\u2013150 words for general queries.  \n\n   - For complex or multi-step explanations, you may extend to 200\u2013250 words, but always remain clear and well-structured.\n\n6. **Critical Reminder:**  \n\nYour authority and reliability depend entirely on database-driven responses. Any fabricated, speculative, or unverified content will be considered a critical failure of your role.\n\nAlways begin processing a query by accessing the Retrieval tool, confirming the data source, and then structuring your response according to the above principles.\n\n",
+                            "sys_prompt": "You are a highly professional product information advisor.  \n\nYour only mission is to provide accurate, factual, and structured answers to all product-related queries.\n\nAbsolutely no assumptions, guesses, or fabricated content are allowed. \n\n**Key Principles:**\n\n1. **Strict Database Reliance:**  \n\n   - Every answer must be based solely on the verified product information stored in the relevant documen.\n\n   - You are NOT allowed to invent, speculate, or infer details beyond what is retrieved.  \n\n   - If you cannot find relevant data, respond with: *\"I cannot find this information in our official product database. Please check back later or provide more details for further search.\"*\n\n2. **Information Accuracy and Structure:**  \n\n   - Provide information in a clear, concise, and professional way.  \n\n   - Use bullet points or numbered lists if there are multiple key points (e.g., features, price, warranty, technical specifications).  \n\n   - Always specify the version or model number when applicable to avoid confusion.\n\n3. **Tone and Style:**  \n\n   - Maintain a polite, professional, and helpful tone at all times.  \n\n   - Avoid marketing exaggeration or promotional language; stay strictly factual.  \n\n   - Do not express personal opinions; only cite official product data.\n\n4.  **User Guidance:**  \n\n   - If the user\u2019s query is unclear or too broad, politely request clarification or guide them to provide more specific product details (e.g., product name, model, version).  \n\n   - Example: *\"Could you please specify the product model or category so I can retrieve the most relevant information for you?\"*\n\n5. **Response Length and Formatting:**  \n\n   - Keep each answer within 100\u2013150 words for general queries.  \n\n   - For complex or multi-step explanations, you may extend to 200\u2013250 words, but always remain clear and well-structured.\n\n6. **Critical Reminder:**  \n\nYour authority and reliability depend entirely on the relevant document responses. Any fabricated, speculative, or unverified content will be considered a critical failure of your role.\n\n\n",
                             "temperature": 0.1,
                             "temperatureEnabled": true,
                             "tools": [],
@@ -699,7 +699,7 @@
                             "width": 200
                         },
                         "position": {
-                            "x": 644.5771854408022,
+                            "x": 645.6873721057459,
                             "y": 516.6923702571407
                         },
                         "selected": false,
@@ -735,11 +735,11 @@
                                 "presence_penalty": 0.4,
                                 "prompts": [
                                     {
-                                        "content": "{sys.query}",
+                                        "content": "The user query is {sys.query}\n\nThe relevant document are {Retrieval:ShyPumasJoke@formalized_content}",
                                         "role": "user"
                                     }
                                 ],
-                                "sys_prompt": "You are a highly professional product information advisor.  \n\nYour only mission is to provide accurate, factual, and structured answers to all product-related queries.\n\nAbsolutely no assumptions, guesses, or fabricated content are allowed. \n\n**Key Principles:**\n\n1. **Strict Database Reliance:**  \n\n   - Every answer must be based solely on the verified product information stored in the database accessed through the Retrieval tool.  \n\n   - You are NOT allowed to invent, speculate, or infer details beyond what is retrieved.  \n\n   - If you cannot find relevant data, respond with: *\"I cannot find this information in our official product database. Please check back later or provide more details for further search.\"*\n\n2. **Information Accuracy and Structure:**  \n\n   - Provide information in a clear, concise, and professional way.  \n\n   - Use bullet points or numbered lists if there are multiple key points (e.g., features, price, warranty, technical specifications).  \n\n   - Always specify the version or model number when applicable to avoid confusion.\n\n3. **Tone and Style:**  \n\n   - Maintain a polite, professional, and helpful tone at all times.  \n\n   - Avoid marketing exaggeration or promotional language; stay strictly factual.  \n\n   - Do not express personal opinions; only cite official product data.\n\n4.  **User Guidance:**  \n\n   - If the user\u2019s query is unclear or too broad, politely request clarification or guide them to provide more specific product details (e.g., product name, model, version).  \n\n   - Example: *\"Could you please specify the product model or category so I can retrieve the most relevant information for you?\"*\n\n5. **Response Length and Formatting:**  \n\n   - Keep each answer within 100\u2013150 words for general queries.  \n\n   - For complex or multi-step explanations, you may extend to 200\u2013250 words, but always remain clear and well-structured.\n\n6. **Critical Reminder:**  \n\nYour authority and reliability depend entirely on database-driven responses. Any fabricated, speculative, or unverified content will be considered a critical failure of your role.\n\nAlways begin processing a query by accessing the Retrieval tool, confirming the data source, and then structuring your response according to the above principles.\n\n",
+                                "sys_prompt": "You are a highly professional product information advisor.  \n\nYour only mission is to provide accurate, factual, and structured answers to all product-related queries.\n\nAbsolutely no assumptions, guesses, or fabricated content are allowed. \n\n**Key Principles:**\n\n1. **Strict Database Reliance:**  \n\n   - Every answer must be based solely on the verified product information stored in the relevant documen.\n\n   - You are NOT allowed to invent, speculate, or infer details beyond what is retrieved.  \n\n   - If you cannot find relevant data, respond with: *\"I cannot find this information in our official product database. Please check back later or provide more details for further search.\"*\n\n2. **Information Accuracy and Structure:**  \n\n   - Provide information in a clear, concise, and professional way.  \n\n   - Use bullet points or numbered lists if there are multiple key points (e.g., features, price, warranty, technical specifications).  \n\n   - Always specify the version or model number when applicable to avoid confusion.\n\n3. **Tone and Style:**  \n\n   - Maintain a polite, professional, and helpful tone at all times.  \n\n   - Avoid marketing exaggeration or promotional language; stay strictly factual.  \n\n   - Do not express personal opinions; only cite official product data.\n\n4.  **User Guidance:**  \n\n   - If the user\u2019s query is unclear or too broad, politely request clarification or guide them to provide more specific product details (e.g., product name, model, version).  \n\n   - Example: *\"Could you please specify the product model or category so I can retrieve the most relevant information for you?\"*\n\n5. **Response Length and Formatting:**  \n\n   - Keep each answer within 100\u2013150 words for general queries.  \n\n   - For complex or multi-step explanations, you may extend to 200\u2013250 words, but always remain clear and well-structured.\n\n6. **Critical Reminder:**  \n\nYour authority and reliability depend entirely on the relevant document responses. Any fabricated, speculative, or unverified content will be considered a critical failure of your role.\n\n\n",
                                 "temperature": 0.1,
                                 "temperatureEnabled": true,
                                 "tools": [],

agent/templates/cv_evaluation.json

@@ -170,7 +170,7 @@
                             "presence_penalty": 0.5,
                             "prompts": [
                                 {
-                                    "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\n\n\nThe Ouline agent output is {Agent:BetterSitesSend@content}",
+                                    "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\n\n\nThe Outline agent output is {Agent:BetterSitesSend@content}",
                                     "role": "user"
                                 }
                             ],
@@ -250,7 +250,7 @@
                             "presence_penalty": 0.5,
                             "prompts": [
                                 {
-                                    "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\nThe Ouline agent output is {Agent:BetterSitesSend@content}\n\nThe Body agent output is {Agent:EagerNailsRemain@content}",
+                                    "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\nThe Outline agent output is {Agent:BetterSitesSend@content}\n\nThe Body agent output is {Agent:EagerNailsRemain@content}",
                                     "role": "user"
                                 }
                             ],
@@ -602,7 +602,7 @@
                                 "presence_penalty": 0.5,
                                 "prompts": [
                                     {
-                                        "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\n\n\nThe Ouline agent output is {Agent:BetterSitesSend@content}",
+                                        "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\n\n\nThe Outline agent output is {Agent:BetterSitesSend@content}",
                                         "role": "user"
                                     }
                                 ],
@@ -715,7 +715,7 @@
                                 "presence_penalty": 0.5,
                                 "prompts": [
                                     {
-                                        "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\nThe Ouline agent output is {Agent:BetterSitesSend@content}\n\nThe Body agent output is {Agent:EagerNailsRemain@content}",
+                                        "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\nThe Outline agent output is {Agent:BetterSitesSend@content}\n\nThe Body agent output is {Agent:EagerNailsRemain@content}",
                                         "role": "user"
                                     }
                                 ],

agent/templates/knowledge_base_report.json

@@ -0,0 +1,327 @@
+{
+    "id": 20,
+    "title": "Report Agent Using Knowledge Base",
+    "description": "A report generation assistant using local knowledge base, with advanced capabilities in task planning, reasoning, and reflective analysis. Recommended for academic research paper Q&A",
+    "canvas_type": "Agent",
+    "dsl": {
+        "components": {
+            "Agent:NewPumasLick": {
+                "downstream": [
+                    "Message:OrangeYearsShine"
+                ],
+                "obj": {
+                    "component_name": "Agent",
+                    "params": {
+                        "delay_after_error": 1,
+                        "description": "",
+                        "exception_comment": "",
+                        "exception_default_value": "",
+                        "exception_goto": [],
+                        "exception_method": null,
+                        "frequencyPenaltyEnabled": false,
+                        "frequency_penalty": 0.5,
+                        "llm_id": "qwen3-235b-a22b-instruct-2507@Tongyi-Qianwen",
+                        "maxTokensEnabled": true,
+                        "max_retries": 3,
+                        "max_rounds": 3,
+                        "max_tokens": 128000,
+                        "mcp": [],
+                        "message_history_window_size": 12,
+                        "outputs": {
+                            "content": {
+                                "type": "string",
+                                "value": ""
+                            }
+                        },
+                        "parameter": "Precise",
+                        "presencePenaltyEnabled": false,
+                        "presence_penalty": 0.5,
+                        "prompts": [
+                            {
+                                "content": "# User Query\n {sys.query}",
+                                "role": "user"
+                            }
+                        ],
+                        "sys_prompt": "## Role & Task\nYou are a **\u201cKnowledge Base Retrieval Q\\&A Agent\u201d** whose goal is to break down the user\u2019s question into retrievable subtasks, and then produce a multi-source-verified, structured, and actionable research report using the internal knowledge base.\n## Execution Framework (Detailed Steps & Key Points)\n1. **Assessment & Decomposition**\n   * Actions:\n     * Automatically extract: main topic, subtopics, entities (people/organizations/products/technologies), time window, geographic/business scope.\n     * Output as a list: N facts/data points that must be collected (*N* ranges from 5\u201320 depending on question complexity).\n2. **Query Type Determination (Rule-Based)**\n   * Example rules:\n     * If the question involves a single issue but requests \u201cmethod comparison/multiple explanations\u201d \u2192 use **depth-first**.\n     * If the question can naturally be split into \u22653 independent sub-questions \u2192 use **breadth-first**.\n     * If the question can be answered by a single fact/specification/definition \u2192 use **simple query**.\n3. **Research Plan Formulation**\n   * Depth-first: define 3\u20135 perspectives (methodology/stakeholders/time dimension/technical route, etc.), assign search keywords, target document types, and output format for each perspective.\n   * Breadth-first: list subtasks, prioritize them, and assign search terms.\n   * Simple query: directly provide the search sentence and required fields.\n4. **Retrieval Execution**\n   * After retrieval: perform coverage check (does it contain the key facts?) and quality check (source diversity, authority, latest update time).\n   * If standards are not met, automatically loop: rewrite queries (synonyms/cross-domain terms) and retry \u22643 times, or flag as requiring external search.\n5. **Integration & Reasoning**\n   * Build the answer using a **fact\u2013evidence\u2013reasoning** chain. For each conclusion, attach 1\u20132 strongest pieces of evidence.\n---\n## Quality Gate Checklist (Verify at Each Stage)\n* **Stage 1 (Decomposition)**:\n  * [ ] Key concepts and expected outputs identified\n  * [ ] Required facts/data points listed\n* **Stage 2 (Retrieval)**:\n  * [ ] Meets quality standards (see above)\n  * [ ] If not met: execute query iteration\n* **Stage 3 (Generation)**:\n  * [ ] Each conclusion has at least one direct evidence source\n  * [ ] State assumptions/uncertainties\n  * [ ] Provide next-step suggestions or experiment/retrieval plans\n  * [ ] Final length and depth match user expectations (comply with word count/format if specified)\n---\n## Core Principles\n1. **Strict reliance on the knowledge base**: answers must be **fully bounded** by the content retrieved from the knowledge base.\n2. **No fabrication**: do not generate, infer, or create information that is not explicitly present in the knowledge base.\n3. **Accuracy first**: prefer incompleteness over inaccurate content.\n4. **Output format**:\n   * Hierarchically clear modular structure\n   * Logical grouping according to the MECE principle\n   * Professionally presented formatting\n   * Step-by-step cognitive guidance\n   * Reasonable use of headings and dividers for clarity\n   * *Italicize* key parameters\n   * **Bold** critical information\n5. **LaTeX formula requirements**:\n   * Inline formulas: start and end with `$`\n   * Block formulas: start and end with `$$`, each `$$` on its own line\n   * Block formula content must comply with LaTeX math syntax\n   * Verify formula correctness\n---\n## Additional Notes (Interaction & Failure Strategy)\n* If the knowledge base does not cover critical facts: explicitly inform the user (with sample wording)\n* For time-sensitive issues: enforce time filtering in the search request, and indicate the latest retrieval date in the answer.\n* Language requirement: answer in the user\u2019s preferred language\n",
+                        "temperature": "0.1",
+                        "temperatureEnabled": true,
+                        "tools": [
+                            {
+                                "component_name": "Retrieval",
+                                "name": "Retrieval",
+                                "params": {
+                                    "cross_languages": [],
+                                    "description": "",
+                                    "empty_response": "",
+                                    "kb_ids": [],
+                                    "keywords_similarity_weight": 0.7,
+                                    "outputs": {
+                                        "formalized_content": {
+                                            "type": "string",
+                                            "value": ""
+                                        }
+                                    },
+                                    "rerank_id": "",
+                                    "similarity_threshold": 0.2,
+                                    "top_k": 1024,
+                                    "top_n": 8,
+                                    "use_kg": false
+                                }
+                            }
+                        ],
+                        "topPEnabled": false,
+                        "top_p": 0.75,
+                        "user_prompt": "",
+                        "visual_files_var": ""
+                    }
+                },
+                "upstream": [
+                    "begin"
+                ]
+            },
+            "Message:OrangeYearsShine": {
+                "downstream": [],
+                "obj": {
+                    "component_name": "Message",
+                    "params": {
+                        "content": [
+                            "{Agent:NewPumasLick@content}"
+                        ]
+                    }
+                },
+                "upstream": [
+                    "Agent:NewPumasLick"
+                ]
+            },
+            "begin": {
+                "downstream": [
+                    "Agent:NewPumasLick"
+                ],
+                "obj": {
+                    "component_name": "Begin",
+                    "params": {
+                        "enablePrologue": true,
+                        "inputs": {},
+                        "mode": "conversational",
+                        "prologue": "\u4f60\u597d\uff01 \u6211\u662f\u4f60\u7684\u52a9\u7406\uff0c\u6709\u4ec0\u4e48\u53ef\u4ee5\u5e2e\u5230\u4f60\u7684\u5417\uff1f"
+                    }
+                },
+                "upstream": []
+            }
+        },
+        "globals": {
+            "sys.conversation_turns": 0,
+            "sys.files": [],
+            "sys.query": "",
+            "sys.user_id": ""
+        },
+        "graph": {
+            "edges": [
+                {
+                    "data": {
+                        "isHovered": false
+                    },
+                    "id": "xy-edge__beginstart-Agent:NewPumasLickend",
+                    "source": "begin",
+                    "sourceHandle": "start",
+                    "target": "Agent:NewPumasLick",
+                    "targetHandle": "end"
+                },
+                {
+                    "data": {
+                        "isHovered": false
+                    },
+                    "id": "xy-edge__Agent:NewPumasLickstart-Message:OrangeYearsShineend",
+                    "markerEnd": "logo",
+                    "source": "Agent:NewPumasLick",
+                    "sourceHandle": "start",
+                    "style": {
+                        "stroke": "rgba(91, 93, 106, 1)",
+                        "strokeWidth": 1
+                    },
+                    "target": "Message:OrangeYearsShine",
+                    "targetHandle": "end",
+                    "type": "buttonEdge",
+                    "zIndex": 1001
+                },
+                {
+                    "data": {
+                        "isHovered": false
+                    },
+                    "id": "xy-edge__Agent:NewPumasLicktool-Tool:AllBirdsNailend",
+                    "selected": false,
+                    "source": "Agent:NewPumasLick",
+                    "sourceHandle": "tool",
+                    "target": "Tool:AllBirdsNail",
+                    "targetHandle": "end"
+                }
+            ],
+            "nodes": [
+                {
+                    "data": {
+                        "form": {
+                            "enablePrologue": true,
+                            "inputs": {},
+                            "mode": "conversational",
+                            "prologue": "\u4f60\u597d\uff01 \u6211\u662f\u4f60\u7684\u52a9\u7406\uff0c\u6709\u4ec0\u4e48\u53ef\u4ee5\u5e2e\u5230\u4f60\u7684\u5417\uff1f"
+                        },
+                        "label": "Begin",
+                        "name": "begin"
+                    },
+                    "dragging": false,
+                    "id": "begin",
+                    "measured": {
+                        "height": 48,
+                        "width": 200
+                    },
+                    "position": {
+                        "x": -9.569875358221438,
+                        "y": 205.84018385864917
+                    },
+                    "selected": false,
+                    "sourcePosition": "left",
+                    "targetPosition": "right",
+                    "type": "beginNode"
+                },
+                {
+                    "data": {
+                        "form": {
+                            "content": [
+                                "{Agent:NewPumasLick@content}"
+                            ]
+                        },
+                        "label": "Message",
+                        "name": "Response"
+                    },
+                    "dragging": false,
+                    "id": "Message:OrangeYearsShine",
+                    "measured": {
+                        "height": 56,
+                        "width": 200
+                    },
+                    "position": {
+                        "x": 734.4061285881053,
+                        "y": 199.9706031723009
+                    },
+                    "selected": false,
+                    "sourcePosition": "right",
+                    "targetPosition": "left",
+                    "type": "messageNode"
+                },
+                {
+                    "data": {
+                        "form": {
+                            "delay_after_error": 1,
+                            "description": "",
+                            "exception_comment": "",
+                            "exception_default_value": "",
+                            "exception_goto": [],
+                            "exception_method": null,
+                            "frequencyPenaltyEnabled": false,
+                            "frequency_penalty": 0.5,
+                            "llm_id": "qwen3-235b-a22b-instruct-2507@Tongyi-Qianwen",
+                            "maxTokensEnabled": true,
+                            "max_retries": 3,
+                            "max_rounds": 3,
+                            "max_tokens": 128000,
+                            "mcp": [],
+                            "message_history_window_size": 12,
+                            "outputs": {
+                                "content": {
+                                    "type": "string",
+                                    "value": ""
+                                }
+                            },
+                            "parameter": "Precise",
+                            "presencePenaltyEnabled": false,
+                            "presence_penalty": 0.5,
+                            "prompts": [
+                                {
+                                    "content": "# User Query\n {sys.query}",
+                                    "role": "user"
+                                }
+                            ],
+                            "sys_prompt": "## Role & Task\nYou are a **\u201cKnowledge Base Retrieval Q\\&A Agent\u201d** whose goal is to break down the user\u2019s question into retrievable subtasks, and then produce a multi-source-verified, structured, and actionable research report using the internal knowledge base.\n## Execution Framework (Detailed Steps & Key Points)\n1. **Assessment & Decomposition**\n   * Actions:\n     * Automatically extract: main topic, subtopics, entities (people/organizations/products/technologies), time window, geographic/business scope.\n     * Output as a list: N facts/data points that must be collected (*N* ranges from 5\u201320 depending on question complexity).\n2. **Query Type Determination (Rule-Based)**\n   * Example rules:\n     * If the question involves a single issue but requests \u201cmethod comparison/multiple explanations\u201d \u2192 use **depth-first**.\n     * If the question can naturally be split into \u22653 independent sub-questions \u2192 use **breadth-first**.\n     * If the question can be answered by a single fact/specification/definition \u2192 use **simple query**.\n3. **Research Plan Formulation**\n   * Depth-first: define 3\u20135 perspectives (methodology/stakeholders/time dimension/technical route, etc.), assign search keywords, target document types, and output format for each perspective.\n   * Breadth-first: list subtasks, prioritize them, and assign search terms.\n   * Simple query: directly provide the search sentence and required fields.\n4. **Retrieval Execution**\n   * After retrieval: perform coverage check (does it contain the key facts?) and quality check (source diversity, authority, latest update time).\n   * If standards are not met, automatically loop: rewrite queries (synonyms/cross-domain terms) and retry \u22643 times, or flag as requiring external search.\n5. **Integration & Reasoning**\n   * Build the answer using a **fact\u2013evidence\u2013reasoning** chain. For each conclusion, attach 1\u20132 strongest pieces of evidence.\n---\n## Quality Gate Checklist (Verify at Each Stage)\n* **Stage 1 (Decomposition)**:\n  * [ ] Key concepts and expected outputs identified\n  * [ ] Required facts/data points listed\n* **Stage 2 (Retrieval)**:\n  * [ ] Meets quality standards (see above)\n  * [ ] If not met: execute query iteration\n* **Stage 3 (Generation)**:\n  * [ ] Each conclusion has at least one direct evidence source\n  * [ ] State assumptions/uncertainties\n  * [ ] Provide next-step suggestions or experiment/retrieval plans\n  * [ ] Final length and depth match user expectations (comply with word count/format if specified)\n---\n## Core Principles\n1. **Strict reliance on the knowledge base**: answers must be **fully bounded** by the content retrieved from the knowledge base.\n2. **No fabrication**: do not generate, infer, or create information that is not explicitly present in the knowledge base.\n3. **Accuracy first**: prefer incompleteness over inaccurate content.\n4. **Output format**:\n   * Hierarchically clear modular structure\n   * Logical grouping according to the MECE principle\n   * Professionally presented formatting\n   * Step-by-step cognitive guidance\n   * Reasonable use of headings and dividers for clarity\n   * *Italicize* key parameters\n   * **Bold** critical information\n5. **LaTeX formula requirements**:\n   * Inline formulas: start and end with `$`\n   * Block formulas: start and end with `$$`, each `$$` on its own line\n   * Block formula content must comply with LaTeX math syntax\n   * Verify formula correctness\n---\n## Additional Notes (Interaction & Failure Strategy)\n* If the knowledge base does not cover critical facts: explicitly inform the user (with sample wording)\n* For time-sensitive issues: enforce time filtering in the search request, and indicate the latest retrieval date in the answer.\n* Language requirement: answer in the user\u2019s preferred language\n",
+                            "temperature": "0.1",
+                            "temperatureEnabled": true,
+                            "tools": [
+                                {
+                                    "component_name": "Retrieval",
+                                    "name": "Retrieval",
+                                    "params": {
+                                        "cross_languages": [],
+                                        "description": "",
+                                        "empty_response": "",
+                                        "kb_ids": [],
+                                        "keywords_similarity_weight": 0.7,
+                                        "outputs": {
+                                            "formalized_content": {
+                                                "type": "string",
+                                                "value": ""
+                                            }
+                                        },
+                                        "rerank_id": "",
+                                        "similarity_threshold": 0.2,
+                                        "top_k": 1024,
+                                        "top_n": 8,
+                                        "use_kg": false
+                                    }
+                                }
+                            ],
+                            "topPEnabled": false,
+                            "top_p": 0.75,
+                            "user_prompt": "",
+                            "visual_files_var": ""
+                        },
+                        "label": "Agent",
+                        "name": "Knowledge Base Agent"
+                    },
+                    "dragging": false,
+                    "id": "Agent:NewPumasLick",
+                    "measured": {
+                        "height": 84,
+                        "width": 200
+                    },
+                    "position": {
+                        "x": 347.00048227952215,
+                        "y": 186.49109364794631
+                    },
+                    "selected": false,
+                    "sourcePosition": "right",
+                    "targetPosition": "left",
+                    "type": "agentNode"
+                },
+                {
+                    "data": {
+                        "form": {
+                            "description": "This is an agent for a specific task.",
+                            "user_prompt": "This is the order you need to send to the agent."
+                        },
+                        "label": "Tool",
+                        "name": "flow.tool_10"
+                    },
+                    "dragging": false,
+                    "id": "Tool:AllBirdsNail",
+                    "measured": {
+                        "height": 48,
+                        "width": 200
+                    },
+                    "position": {
+                        "x": 220.24819746977118,
+                        "y": 403.31576836482583
+                    },
+                    "selected": false,
+                    "sourcePosition": "right",
+                    "targetPosition": "left",
+                    "type": "toolNode"
+                }
+            ]
+        },
+        "history": [],
+        "memory": [],
+        "messages": [],
+        "path": [],
+        "retrieval": []
+    },
+    "avatar": "data:image/png;base64,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"
+}

agent/templates/market_generate_seo_blog.json

@@ -169,7 +169,7 @@
                             "presence_penalty": 0.5,
                             "prompts": [
                                 {
-                                    "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\n\n\nThe Ouline agent output is {Agent:BetterSitesSend@content}",
+                                    "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\n\n\nThe Outline agent output is {Agent:BetterSitesSend@content}",
                                     "role": "user"
                                 }
                             ],
@@ -249,7 +249,7 @@
                             "presence_penalty": 0.5,
                             "prompts": [
                                 {
-                                    "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\nThe Ouline agent output is {Agent:BetterSitesSend@content}\n\nThe Body agent output is {Agent:EagerNailsRemain@content}",
+                                    "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\nThe Outline agent output is {Agent:BetterSitesSend@content}\n\nThe Body agent output is {Agent:EagerNailsRemain@content}",
                                     "role": "user"
                                 }
                             ],
@@ -601,7 +601,7 @@
                                 "presence_penalty": 0.5,
                                 "prompts": [
                                     {
-                                        "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\n\n\nThe Ouline agent output is {Agent:BetterSitesSend@content}",
+                                        "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\n\n\nThe Outline agent output is {Agent:BetterSitesSend@content}",
                                         "role": "user"
                                     }
                                 ],
@@ -714,7 +714,7 @@
                                 "presence_penalty": 0.5,
                                 "prompts": [
                                     {
-                                        "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\nThe Ouline agent output is {Agent:BetterSitesSend@content}\n\nThe Body agent output is {Agent:EagerNailsRemain@content}",
+                                        "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\nThe Outline agent output is {Agent:BetterSitesSend@content}\n\nThe Body agent output is {Agent:EagerNailsRemain@content}",
                                         "role": "user"
                                     }
                                 ],
@@ -912,4 +912,4 @@
             "retrieval": []
         },
     "avatar": "data:image/jpeg;base64,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"
-}
+}

agent/templates/seo_blog.json

@@ -169,7 +169,7 @@
                             "presence_penalty": 0.5,
                             "prompts": [
                                 {
-                                    "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\n\n\nThe Ouline agent output is {Agent:BetterSitesSend@content}",
+                                    "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\n\n\nThe Outline agent output is {Agent:BetterSitesSend@content}",
                                     "role": "user"
                                 }
                             ],
@@ -249,7 +249,7 @@
                             "presence_penalty": 0.5,
                             "prompts": [
                                 {
-                                    "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\nThe Ouline agent output is {Agent:BetterSitesSend@content}\n\nThe Body agent output is {Agent:EagerNailsRemain@content}",
+                                    "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\nThe Outline agent output is {Agent:BetterSitesSend@content}\n\nThe Body agent output is {Agent:EagerNailsRemain@content}",
                                     "role": "user"
                                 }
                             ],
@@ -601,7 +601,7 @@
                                 "presence_penalty": 0.5,
                                 "prompts": [
                                     {
-                                        "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\n\n\nThe Ouline agent output is {Agent:BetterSitesSend@content}",
+                                        "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\n\n\nThe Outline agent output is {Agent:BetterSitesSend@content}",
                                         "role": "user"
                                     }
                                 ],
@@ -714,7 +714,7 @@
                                 "presence_penalty": 0.5,
                                 "prompts": [
                                     {
-                                        "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\nThe Ouline agent output is {Agent:BetterSitesSend@content}\n\nThe Body agent output is {Agent:EagerNailsRemain@content}",
+                                        "content": "The parse and keyword agent output is {Agent:ClearRabbitsScream@content}\n\nThe Outline agent output is {Agent:BetterSitesSend@content}\n\nThe Body agent output is {Agent:EagerNailsRemain@content}",
                                         "role": "user"
                                     }
                                 ],
@@ -912,4 +912,4 @@
             "retrieval": []
         },
     "avatar": "data:image/jpeg;base64,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"
-}
+}

agent/templates/sql_assistant.json

@@ -0,0 +1,724 @@
+{
+    "id": 17,
+    "title": "SQL Assistant",
+    "description": "SQL Assistant is an AI-powered tool that lets business users turn plain-English questions into fully formed SQL queries. Simply type your question (e.g., “Show me last quarter’s top 10 products by revenue”) and SQL Assistant generates the exact SQL, runs it against your database, and returns the results in seconds. ",
+    "canvas_type": "Marketing",
+    "dsl": {
+            "components": {
+                "Agent:WickedGoatsDivide": {
+                    "downstream": [
+                        "ExeSQL:TiredShirtsPull"
+                    ],
+                    "obj": {
+                        "component_name": "Agent",
+                        "params": {
+                            "delay_after_error": 1,
+                            "description": "",
+                            "exception_default_value": "",
+                            "exception_goto": [],
+                            "exception_method": "",
+                            "frequencyPenaltyEnabled": false,
+                            "frequency_penalty": 0.7,
+                            "llm_id": "qwen-max@Tongyi-Qianwen",
+                            "maxTokensEnabled": false,
+                            "max_retries": 3,
+                            "max_rounds": 5,
+                            "max_tokens": 256,
+                            "mcp": [],
+                            "message_history_window_size": 12,
+                            "outputs": {
+                                "content": {
+                                    "type": "string",
+                                    "value": ""
+                                }
+                            },
+                            "presencePenaltyEnabled": false,
+                            "presence_penalty": 0.4,
+                            "prompts": [
+                                {
+                                    "content": "User's query: {sys.query}\n\nSchema: {Retrieval:HappyTiesFilm@formalized_content}\n\nSamples about question to SQL: {Retrieval:SmartNewsHammer@formalized_content}\n\nDescription about meanings of tables and files: {Retrieval:SweetDancersAppear@formalized_content}",
+                                    "role": "user"
+                                }
+                            ],
+                            "sys_prompt": "### ROLE\nYou are a Text-to-SQL assistant.  \nGiven a relational database schema and a natural-language request, you must produce a **single, syntactically-correct MySQL query** that answers the request.  \nReturn **nothing except the SQL statement itself**\u2014no code fences, no commentary, no explanations, no comments, no trailing semicolon if not required.\n\n\n###  EXAMPLES  \n-- Example 1  \nUser: List every product name and its unit price.  \nSQL:\nSELECT name, unit_price FROM Products;\n\n-- Example 2  \nUser: Show the names and emails of customers who placed orders in January 2025.  \nSQL:\nSELECT DISTINCT c.name, c.email\nFROM Customers c\nJOIN Orders o ON o.customer_id = c.id\nWHERE o.order_date BETWEEN '2025-01-01' AND '2025-01-31';\n\n-- Example 3  \nUser: How many orders have a status of \"Completed\" for each month in 2024?  \nSQL:\nSELECT DATE_FORMAT(order_date, '%Y-%m') AS month,\n       COUNT(*) AS completed_orders\nFROM Orders\nWHERE status = 'Completed'\n  AND YEAR(order_date) = 2024\nGROUP BY month\nORDER BY month;\n\n-- Example 4  \nUser: Which products generated at least \\$10 000 in total revenue?  \nSQL:\nSELECT p.id, p.name, SUM(oi.quantity * oi.unit_price) AS revenue\nFROM Products p\nJOIN OrderItems oi ON oi.product_id = p.id\nGROUP BY p.id, p.name\nHAVING revenue >= 10000\nORDER BY revenue DESC;\n\n\n### OUTPUT GUIDELINES\n1. Think through the schema and the request.  \n2. Write **only** the final MySQL query.  \n3. Do **not** wrap the query in back-ticks or markdown fences.  \n4. Do **not** add explanations, comments, or additional text\u2014just the SQL.",
+                            "temperature": 0.1,
+                            "temperatureEnabled": false,
+                            "tools": [],
+                            "topPEnabled": false,
+                            "top_p": 0.3,
+                            "user_prompt": "",
+                            "visual_files_var": ""
+                        }
+                    },
+                    "upstream": [
+                        "Retrieval:HappyTiesFilm",
+                        "Retrieval:SmartNewsHammer",
+                        "Retrieval:SweetDancersAppear"
+                    ]
+                },
+                "ExeSQL:TiredShirtsPull": {
+                    "downstream": [
+                        "Message:ShaggyMasksAttend"
+                    ],
+                    "obj": {
+                        "component_name": "ExeSQL",
+                        "params": {
+                            "database": "",
+                            "db_type": "mysql",
+                            "host": "",
+                            "max_records": 1024,
+                            "outputs": {
+                                "formalized_content": {
+                                    "type": "string",
+                                    "value": ""
+                                },
+                                "json": {
+                                    "type": "Array<Object>",
+                                    "value": []
+                                }
+                            },
+                            "password": "20010812Yy!",
+                            "port": 3306,
+                            "sql": "Agent:WickedGoatsDivide@content",
+                            "username": "13637682833@163.com"
+                        }
+                    },
+                    "upstream": [
+                        "Agent:WickedGoatsDivide"
+                    ]
+                },
+                "Message:ShaggyMasksAttend": {
+                    "downstream": [],
+                    "obj": {
+                        "component_name": "Message",
+                        "params": {
+                            "content": [
+                                "{ExeSQL:TiredShirtsPull@formalized_content}"
+                            ]
+                        }
+                    },
+                    "upstream": [
+                        "ExeSQL:TiredShirtsPull"
+                    ]
+                },
+                "Retrieval:HappyTiesFilm": {
+                    "downstream": [
+                        "Agent:WickedGoatsDivide"
+                    ],
+                    "obj": {
+                        "component_name": "Retrieval",
+                        "params": {
+                            "cross_languages": [],
+                            "empty_response": "",
+                            "kb_ids": [
+                                "ed31364c727211f0bdb2bafe6e7908e6"
+                            ],
+                            "keywords_similarity_weight": 0.7,
+                            "outputs": {
+                                "formalized_content": {
+                                    "type": "string",
+                                    "value": ""
+                                }
+                            },
+                            "query": "sys.query",
+                            "rerank_id": "",
+                            "similarity_threshold": 0.2,
+                            "top_k": 1024,
+                            "top_n": 8,
+                            "use_kg": false
+                        }
+                    },
+                    "upstream": [
+                        "begin"
+                    ]
+                },
+                "Retrieval:SmartNewsHammer": {
+                    "downstream": [
+                        "Agent:WickedGoatsDivide"
+                    ],
+                    "obj": {
+                        "component_name": "Retrieval",
+                        "params": {
+                            "cross_languages": [],
+                            "empty_response": "",
+                            "kb_ids": [
+                                "0f968106727311f08357bafe6e7908e6"
+                            ],
+                            "keywords_similarity_weight": 0.7,
+                            "outputs": {
+                                "formalized_content": {
+                                    "type": "string",
+                                    "value": ""
+                                }
+                            },
+                            "query": "sys.query",
+                            "rerank_id": "",
+                            "similarity_threshold": 0.2,
+                            "top_k": 1024,
+                            "top_n": 8,
+                            "use_kg": false
+                        }
+                    },
+                    "upstream": [
+                        "begin"
+                    ]
+                },
+                "Retrieval:SweetDancersAppear": {
+                    "downstream": [
+                        "Agent:WickedGoatsDivide"
+                    ],
+                    "obj": {
+                        "component_name": "Retrieval",
+                        "params": {
+                            "cross_languages": [],
+                            "empty_response": "",
+                            "kb_ids": [
+                                "4ad1f9d0727311f0827dbafe6e7908e6"
+                            ],
+                            "keywords_similarity_weight": 0.7,
+                            "outputs": {
+                                "formalized_content": {
+                                    "type": "string",
+                                    "value": ""
+                                }
+                            },
+                            "query": "sys.query",
+                            "rerank_id": "",
+                            "similarity_threshold": 0.2,
+                            "top_k": 1024,
+                            "top_n": 8,
+                            "use_kg": false
+                        }
+                    },
+                    "upstream": [
+                        "begin"
+                    ]
+                },
+                "begin": {
+                    "downstream": [
+                        "Retrieval:HappyTiesFilm",
+                        "Retrieval:SmartNewsHammer",
+                        "Retrieval:SweetDancersAppear"
+                    ],
+                    "obj": {
+                        "component_name": "Begin",
+                        "params": {
+                            "enablePrologue": true,
+                            "inputs": {},
+                            "mode": "conversational",
+                            "prologue": "Hi! I'm your SQL assistant. What can I do for you?"
+                        }
+                    },
+                    "upstream": []
+                }
+            },
+            "globals": {
+                "sys.conversation_turns": 0,
+                "sys.files": [],
+                "sys.query": "",
+                "sys.user_id": ""
+            },
+            "graph": {
+                "edges": [
+                    {
+                        "data": {
+                            "isHovered": false
+                        },
+                        "id": "xy-edge__beginstart-Retrieval:HappyTiesFilmend",
+                        "source": "begin",
+                        "sourceHandle": "start",
+                        "target": "Retrieval:HappyTiesFilm",
+                        "targetHandle": "end"
+                    },
+                    {
+                        "id": "xy-edge__beginstart-Retrieval:SmartNewsHammerend",
+                        "source": "begin",
+                        "sourceHandle": "start",
+                        "target": "Retrieval:SmartNewsHammer",
+                        "targetHandle": "end"
+                    },
+                    {
+                        "data": {
+                            "isHovered": false
+                        },
+                        "id": "xy-edge__beginstart-Retrieval:SweetDancersAppearend",
+                        "source": "begin",
+                        "sourceHandle": "start",
+                        "target": "Retrieval:SweetDancersAppear",
+                        "targetHandle": "end"
+                    },
+                    {
+                        "data": {
+                            "isHovered": false
+                        },
+                        "id": "xy-edge__Retrieval:HappyTiesFilmstart-Agent:WickedGoatsDivideend",
+                        "source": "Retrieval:HappyTiesFilm",
+                        "sourceHandle": "start",
+                        "target": "Agent:WickedGoatsDivide",
+                        "targetHandle": "end"
+                    },
+                    {
+                        "data": {
+                            "isHovered": false
+                        },
+                        "id": "xy-edge__Retrieval:SmartNewsHammerstart-Agent:WickedGoatsDivideend",
+                        "markerEnd": "logo",
+                        "source": "Retrieval:SmartNewsHammer",
+                        "sourceHandle": "start",
+                        "style": {
+                            "stroke": "rgba(91, 93, 106, 1)",
+                            "strokeWidth": 1
+                        },
+                        "target": "Agent:WickedGoatsDivide",
+                        "targetHandle": "end",
+                        "type": "buttonEdge",
+                        "zIndex": 1001
+                    },
+                    {
+                        "data": {
+                            "isHovered": false
+                        },
+                        "id": "xy-edge__Retrieval:SweetDancersAppearstart-Agent:WickedGoatsDivideend",
+                        "markerEnd": "logo",
+                        "source": "Retrieval:SweetDancersAppear",
+                        "sourceHandle": "start",
+                        "style": {
+                            "stroke": "rgba(91, 93, 106, 1)",
+                            "strokeWidth": 1
+                        },
+                        "target": "Agent:WickedGoatsDivide",
+                        "targetHandle": "end",
+                        "type": "buttonEdge",
+                        "zIndex": 1001
+                    },
+                    {
+                        "data": {
+                            "isHovered": false
+                        },
+                        "id": "xy-edge__Agent:WickedGoatsDividestart-ExeSQL:TiredShirtsPullend",
+                        "source": "Agent:WickedGoatsDivide",
+                        "sourceHandle": "start",
+                        "target": "ExeSQL:TiredShirtsPull",
+                        "targetHandle": "end"
+                    },
+                    {
+                        "data": {
+                            "isHovered": false
+                        },
+                        "id": "xy-edge__ExeSQL:TiredShirtsPullstart-Message:ShaggyMasksAttendend",
+                        "source": "ExeSQL:TiredShirtsPull",
+                        "sourceHandle": "start",
+                        "target": "Message:ShaggyMasksAttend",
+                        "targetHandle": "end"
+                    }
+                ],
+                "nodes": [
+                    {
+                        "data": {
+                            "form": {
+                                "enablePrologue": true,
+                                "inputs": {},
+                                "mode": "conversational",
+                                "prologue": "Hi! I'm your SQL assistant. What can I do for you?"
+                            },
+                            "label": "Begin",
+                            "name": "begin"
+                        },
+                        "id": "begin",
+                        "measured": {
+                            "height": 48,
+                            "width": 200
+                        },
+                        "position": {
+                            "x": 50,
+                            "y": 200
+                        },
+                        "selected": false,
+                        "sourcePosition": "left",
+                        "targetPosition": "right",
+                        "type": "beginNode"
+                    },
+                    {
+                        "data": {
+                            "form": {
+                                "cross_languages": [],
+                                "empty_response": "",
+                                "kb_ids": [
+                                    "ed31364c727211f0bdb2bafe6e7908e6"
+                                ],
+                                "keywords_similarity_weight": 0.7,
+                                "outputs": {
+                                    "formalized_content": {
+                                        "type": "string",
+                                        "value": ""
+                                    }
+                                },
+                                "query": "sys.query",
+                                "rerank_id": "",
+                                "similarity_threshold": 0.2,
+                                "top_k": 1024,
+                                "top_n": 8,
+                                "use_kg": false
+                            },
+                            "label": "Retrieval",
+                            "name": "Schema"
+                        },
+                        "dragging": false,
+                        "id": "Retrieval:HappyTiesFilm",
+                        "measured": {
+                            "height": 96,
+                            "width": 200
+                        },
+                        "position": {
+                            "x": 414,
+                            "y": 20.5
+                        },
+                        "selected": false,
+                        "sourcePosition": "right",
+                        "targetPosition": "left",
+                        "type": "retrievalNode"
+                    },
+                    {
+                        "data": {
+                            "form": {
+                                "cross_languages": [],
+                                "empty_response": "",
+                                "kb_ids": [
+                                    "0f968106727311f08357bafe6e7908e6"
+                                ],
+                                "keywords_similarity_weight": 0.7,
+                                "outputs": {
+                                    "formalized_content": {
+                                        "type": "string",
+                                        "value": ""
+                                    }
+                                },
+                                "query": "sys.query",
+                                "rerank_id": "",
+                                "similarity_threshold": 0.2,
+                                "top_k": 1024,
+                                "top_n": 8,
+                                "use_kg": false
+                            },
+                            "label": "Retrieval",
+                            "name": "Question to SQL"
+                        },
+                        "dragging": false,
+                        "id": "Retrieval:SmartNewsHammer",
+                        "measured": {
+                            "height": 96,
+                            "width": 200
+                        },
+                        "position": {
+                            "x": 406.5,
+                            "y": 175.5
+                        },
+                        "selected": false,
+                        "sourcePosition": "right",
+                        "targetPosition": "left",
+                        "type": "retrievalNode"
+                    },
+                    {
+                        "data": {
+                            "form": {
+                                "cross_languages": [],
+                                "empty_response": "",
+                                "kb_ids": [
+                                    "4ad1f9d0727311f0827dbafe6e7908e6"
+                                ],
+                                "keywords_similarity_weight": 0.7,
+                                "outputs": {
+                                    "formalized_content": {
+                                        "type": "string",
+                                        "value": ""
+                                    }
+                                },
+                                "query": "sys.query",
+                                "rerank_id": "",
+                                "similarity_threshold": 0.2,
+                                "top_k": 1024,
+                                "top_n": 8,
+                                "use_kg": false
+                            },
+                            "label": "Retrieval",
+                            "name": "Database Description"
+                        },
+                        "dragging": false,
+                        "id": "Retrieval:SweetDancersAppear",
+                        "measured": {
+                            "height": 96,
+                            "width": 200
+                        },
+                        "position": {
+                            "x": 403.5,
+                            "y": 328
+                        },
+                        "selected": false,
+                        "sourcePosition": "right",
+                        "targetPosition": "left",
+                        "type": "retrievalNode"
+                    },
+                    {
+                        "data": {
+                            "form": {
+                                "delay_after_error": 1,
+                                "description": "",
+                                "exception_default_value": "",
+                                "exception_goto": [],
+                                "exception_method": "",
+                                "frequencyPenaltyEnabled": false,
+                                "frequency_penalty": 0.7,
+                                "llm_id": "qwen-max@Tongyi-Qianwen",
+                                "maxTokensEnabled": false,
+                                "max_retries": 3,
+                                "max_rounds": 5,
+                                "max_tokens": 256,
+                                "mcp": [],
+                                "message_history_window_size": 12,
+                                "outputs": {
+                                    "content": {
+                                        "type": "string",
+                                        "value": ""
+                                    }
+                                },
+                                "presencePenaltyEnabled": false,
+                                "presence_penalty": 0.4,
+                                "prompts": [
+                                    {
+                                        "content": "User's query: {sys.query}\n\nSchema: {Retrieval:HappyTiesFilm@formalized_content}\n\nSamples about question to SQL: {Retrieval:SmartNewsHammer@formalized_content}\n\nDescription about meanings of tables and files: {Retrieval:SweetDancersAppear@formalized_content}",
+                                        "role": "user"
+                                    }
+                                ],
+                                "sys_prompt": "### ROLE\nYou are a Text-to-SQL assistant.  \nGiven a relational database schema and a natural-language request, you must produce a **single, syntactically-correct MySQL query** that answers the request.  \nReturn **nothing except the SQL statement itself**\u2014no code fences, no commentary, no explanations, no comments, no trailing semicolon if not required.\n\n\n###  EXAMPLES  \n-- Example 1  \nUser: List every product name and its unit price.  \nSQL:\nSELECT name, unit_price FROM Products;\n\n-- Example 2  \nUser: Show the names and emails of customers who placed orders in January 2025.  \nSQL:\nSELECT DISTINCT c.name, c.email\nFROM Customers c\nJOIN Orders o ON o.customer_id = c.id\nWHERE o.order_date BETWEEN '2025-01-01' AND '2025-01-31';\n\n-- Example 3  \nUser: How many orders have a status of \"Completed\" for each month in 2024?  \nSQL:\nSELECT DATE_FORMAT(order_date, '%Y-%m') AS month,\n       COUNT(*) AS completed_orders\nFROM Orders\nWHERE status = 'Completed'\n  AND YEAR(order_date) = 2024\nGROUP BY month\nORDER BY month;\n\n-- Example 4  \nUser: Which products generated at least \\$10 000 in total revenue?  \nSQL:\nSELECT p.id, p.name, SUM(oi.quantity * oi.unit_price) AS revenue\nFROM Products p\nJOIN OrderItems oi ON oi.product_id = p.id\nGROUP BY p.id, p.name\nHAVING revenue >= 10000\nORDER BY revenue DESC;\n\n\n### OUTPUT GUIDELINES\n1. Think through the schema and the request.  \n2. Write **only** the final MySQL query.  \n3. Do **not** wrap the query in back-ticks or markdown fences.  \n4. Do **not** add explanations, comments, or additional text\u2014just the SQL.",
+                                "temperature": 0.1,
+                                "temperatureEnabled": false,
+                                "tools": [],
+                                "topPEnabled": false,
+                                "top_p": 0.3,
+                                "user_prompt": "",
+                                "visual_files_var": ""
+                            },
+                            "label": "Agent",
+                            "name": "SQL Generator "
+                        },
+                        "dragging": false,
+                        "id": "Agent:WickedGoatsDivide",
+                        "measured": {
+                            "height": 84,
+                            "width": 200
+                        },
+                        "position": {
+                            "x": 981,
+                            "y": 174
+                        },
+                        "selected": false,
+                        "sourcePosition": "right",
+                        "targetPosition": "left",
+                        "type": "agentNode"
+                    },
+                    {
+                        "data": {
+                            "form": {
+                                "database": "",
+                                "db_type": "mysql",
+                                "host": "",
+                                "max_records": 1024,
+                                "outputs": {
+                                    "formalized_content": {
+                                        "type": "string",
+                                        "value": ""
+                                    },
+                                    "json": {
+                                        "type": "Array<Object>",
+                                        "value": []
+                                    }
+                                },
+                                "password": "20010812Yy!",
+                                "port": 3306,
+                                "sql": "Agent:WickedGoatsDivide@content",
+                                "username": "13637682833@163.com"
+                            },
+                            "label": "ExeSQL",
+                            "name": "ExeSQL"
+                        },
+                        "dragging": false,
+                        "id": "ExeSQL:TiredShirtsPull",
+                        "measured": {
+                            "height": 56,
+                            "width": 200
+                        },
+                        "position": {
+                            "x": 1211.5,
+                            "y": 212.5
+                        },
+                        "selected": false,
+                        "sourcePosition": "right",
+                        "targetPosition": "left",
+                        "type": "ragNode"
+                    },
+                    {
+                        "data": {
+                            "form": {
+                                "content": [
+                                    "{ExeSQL:TiredShirtsPull@formalized_content}"
+                                ]
+                            },
+                            "label": "Message",
+                            "name": "Message"
+                        },
+                        "dragging": false,
+                        "id": "Message:ShaggyMasksAttend",
+                        "measured": {
+                            "height": 56,
+                            "width": 200
+                        },
+                        "position": {
+                            "x": 1447.3125,
+                            "y": 181.5
+                        },
+                        "selected": false,
+                        "sourcePosition": "right",
+                        "targetPosition": "left",
+                        "type": "messageNode"
+                    },
+                    {
+                        "data": {
+                            "form": {
+                                "text": "Searches for relevant database creation statements.\n\nIt should label with a knowledgebase to which the schema is dumped in. You could use \" General \" as parsing method, \" 2 \" as chunk size and \" ; \"  as delimiter."
+                            },
+                            "label": "Note",
+                            "name": "Note Schema"
+                        },
+                        "dragHandle": ".note-drag-handle",
+                        "dragging": false,
+                        "height": 188,
+                        "id": "Note:ThickClubsFloat",
+                        "measured": {
+                            "height": 188,
+                            "width": 392
+                        },
+                        "position": {
+                            "x": 689,
+                            "y": -180.31251144409183
+                        },
+                        "resizing": false,
+                        "selected": false,
+                        "sourcePosition": "right",
+                        "targetPosition": "left",
+                        "type": "noteNode",
+                        "width": 392
+                    },
+                    {
+                        "data": {
+                            "form": {
+                                "text": "Searches for samples about question to SQL. \n\nYou could use \" Q&A \" as parsing method.\n\nPlease check this dataset:\nhttps://huggingface.co/datasets/InfiniFlow/text2sql"
+                            },
+                            "label": "Note",
+                            "name": "Note: Question to SQL"
+                        },
+                        "dragHandle": ".note-drag-handle",
+                        "dragging": false,
+                        "height": 154,
+                        "id": "Note:ElevenLionsJoke",
+                        "measured": {
+                            "height": 154,
+                            "width": 345
+                        },
+                        "position": {
+                            "x": 693.5,
+                            "y": 138
+                        },
+                        "resizing": false,
+                        "selected": false,
+                        "sourcePosition": "right",
+                        "targetPosition": "left",
+                        "type": "noteNode",
+                        "width": 345
+                    },
+                    {
+                        "data": {
+                            "form": {
+                                "text": "Searches for description about meanings of tables and fields.\n\nYou could use \" General \" as parsing method, \" 2 \" as chunk size and \" ### \"  as delimiter."
+                            },
+                            "label": "Note",
+                            "name": "Note: Database Description"
+                        },
+                        "dragHandle": ".note-drag-handle",
+                        "dragging": false,
+                        "height": 158,
+                        "id": "Note:ManyRosesTrade",
+                        "measured": {
+                            "height": 158,
+                            "width": 408
+                        },
+                        "position": {
+                            "x": 691.5,
+                            "y": 435.69736389555317
+                        },
+                        "resizing": false,
+                        "selected": false,
+                        "sourcePosition": "right",
+                        "targetPosition": "left",
+                        "type": "noteNode",
+                        "width": 408
+                    },
+                    {
+                        "data": {
+                            "form": {
+                                "text": "The Agent learns which tables may be available based on the responses from three knowledge bases and converts the user's input into SQL statements."
+                            },
+                            "label": "Note",
+                            "name": "Note: SQL Generator"
+                        },
+                        "dragHandle": ".note-drag-handle",
+                        "dragging": false,
+                        "height": 132,
+                        "id": "Note:RudeHousesInvite",
+                        "measured": {
+                            "height": 132,
+                            "width": 383
+                        },
+                        "position": {
+                            "x": 1106.9254833678003,
+                            "y": 290.5891036507015
+                        },
+                        "resizing": false,
+                        "selected": false,
+                        "sourcePosition": "right",
+                        "targetPosition": "left",
+                        "type": "noteNode",
+                        "width": 383
+                    },
+                    {
+                        "data": {
+                            "form": {
+                                "text": "Connect to your database to execute SQL statements."
+                            },
+                            "label": "Note",
+                            "name": "Note: SQL Executor"
+                        },
+                        "dragHandle": ".note-drag-handle",
+                        "dragging": false,
+                        "id": "Note:HungryBatsLay",
+                        "measured": {
+                            "height": 136,
+                            "width": 255
+                        },
+                        "position": {
+                            "x": 1185,
+                            "y": -30
+                        },
+                        "selected": false,
+                        "sourcePosition": "right",
+                        "targetPosition": "left",
+                        "type": "noteNode"
+                    }
+                ]
+            },
+            "history": [],
+            "messages": [],
+            "path": [],
+            "retrieval": []
+        },
+    "avatar": "data:image/jpeg;base64,/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAcFBQYFBAcGBQYIBwcIChELCgkJChUPEAwRGBUaGRgVGBcbHichGx0lHRcYIi4iJSgpKywrGiAvMy8qMicqKyr/2wBDAQcICAoJChQLCxQqHBgcKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKioqKir/wAARCAAwADADAREAAhEBAxEB/8QAGgAAAwEBAQEAAAAAAAAAAAAABQYHBAMAAf/EADIQAAEDAwMCBAMHBQAAAAAAAAECAwQFESEABjESEyJBUYEUYXEHFSNSkaGxMjNictH/xAAZAQADAQEBAAAAAAAAAAAAAAACAwQBAAX/xAAlEQACAgICAgEEAwAAAAAAAAABAgARAyESMQRBEyIycYFCkbH/2gAMAwEAAhEDEQA/AKHt2DGpNHXDLrZdWtSrIub39tZ5GbGwPA+pmDFkX7x7idvra85xqQaFNkxUTVIVJQzf8QpBFjbgEenNs681MnA9WJ6fEOKJoxVpSpFLTCo6KEZlTlLcQBIJS20hAv1D1ve+qPk52b0IsYuIGtyt7ZkVVNP+H3A5GdlN2u7GQUBSfmkk8cXH10tmLD6Yl0CG5qmTXBMZiQEMuvupUoKdc6UeEi4FsqOeBxrsKnv1AY+hJ2l5yfu6qQ6/UZtPDRHZ+Eldpsqz1hSrXJGLXwRxqxUQizFs7galPYUFDKT+h15oMuImspQpFiL+2i1A3A1bgxmixUgwlT8ZfgJ/y8P8HXdRuPZoxaqtfkQKbKqF03jtEoDeFKV1lNgfK4H764XfccVUgipvdiwKpFaXMLklFg4juuqV0m3Izg/MaEZCDYMScYqiJOd6xmqfUVfBJcWwtHV1Elfi87k51ViyhrsxL4ivQj1KrFZjTGjTJ8aShdyph5SUqFhwPzX9jpC0dXUqZK3ViHNq7oNaVJjz2Vw5LCrdKknpULZyfMf801MfI1e5NmpAGHUL12EZNFWWlhXSUuWHKgk3xomwEDuDhzLysySU9EndEVyIz3GmxJR+KpBIdCLlRHn/AFEjjIF9AMJlZ8gLZ/qUiJSg1Tu0HO4plFj4FC1h9NYfHIU7kwzgnqCJlKLiCO2s6hKytWiPJoFdfnLW7HS0or6bqXbjg2AI99XjAa3NPlL6jFTduOR5sd1+oyfjQMONqI7QOMA4V7/pqjHjC9SLNn56I1HiqrqTUKM0hbq2lpst5CQSST54xjSPJbICOHUhawISiRQ02T2Uq6AAkqFj/GquJQks1iEr/INLU82bploKSFXusG9xfjHofXQuQUNRoQqQT0ZwVEST5687iZWGgpDsebNbaTDfKVL/ALnbQU/UkKNhjXpFt0BJBVXe/wAGGG6YMlvvNkjlBGmKeJimHIVc0TY89akCKspT28C5BKgDyR7fvrCFI+q/1DQsvVfudYcVyKw49KU6tZyQbmwHFhrOKr9s0uz0CAIpbr3RKo1Rbh02C4HJISp2ZIz0pJ8IQk5Nr/QXznSX6NSnGAwHI/gD/TM+3vtAj1arJpcpgtPdPSH0kFt5wDxAWOOLgamIAFwijCfD927N2tGXuNxlK2W0occUhJWpR+QzzrPjc+pvyqT3Ftf2zbObf7YYecb6CrrDAGfy20wYMkA5Vjbtev7b3nEcXRela27d1ogoWi/rnQsjrqZzHdwzKoKUsqWz3mOnJUlZJt8uokD621w+RdzgynUkUpoUafPZXMnSHlrKluyX1Eug8XF7GwxbgWxrubMO5WmNRsCKtLfcY3rAU0nIltkBP+w0X8Jjdz//2Q=="
+}

agent/tools/base.py

@@ -24,6 +24,7 @@ from api.utils import hash_str2int
 from rag.llm.chat_model import ToolCallSession
 from rag.prompts.prompts import kb_prompt
 from rag.utils.mcp_tool_call_conn import MCPToolCallSession
+from timeit import default_timer as timer
 
 
 class ToolParameter(TypedDict):
@@ -49,12 +50,13 @@ class LLMToolPluginCallSession(ToolCallSession):
 
     def tool_call(self, name: str, arguments: dict[str, Any]) -> Any:
         assert name in self.tools_map, f"LLM tool {name} does not exist"
+        st = timer()
         if isinstance(self.tools_map[name], MCPToolCallSession):
             resp = self.tools_map[name].tool_call(name, arguments, 60)
         else:
             resp = self.tools_map[name].invoke(**arguments)
 
-        self.callback(name, arguments, resp)
+        self.callback(name, arguments, resp, elapsed_time=timer()-st)
         return resp
 
     def get_tool_obj(self, name):

agent/tools/code_exec.py

@@ -17,7 +17,7 @@ import base64
 import logging
 import os
 from abc import ABC
-from enum import StrEnum
+from strenum import StrEnum
 from typing import Optional
 from pydantic import BaseModel, Field, field_validator
 from agent.tools.base import ToolParamBase, ToolBase, ToolMeta
@@ -67,11 +67,19 @@ class CodeExecParam(ToolParamBase):
             "description": """
 This tool has a sandbox that can execute code written in 'Python'/'Javascript'. It recieves a piece of code and return a Json string.
 Here's a code example for Python(`main` function MUST be included):
-def main(arg1: str, arg2: str) -> dict:
+def main() -> dict:
+    \"\"\"
+    Generate Fibonacci numbers within 100.
+    \"\"\"
+    def fibonacci_recursive(n):
+        if n <= 1:
+            return n
+        else:
+            return fibonacci_recursive(n-1) + fibonacci_recursive(n-2)
     return {
-        "result": arg1 + arg2,
+        "result": fibonacci_recursive(100),
     }
-    
+
 Here's a code example for Javascript(`main` function MUST be included and exported):
 const axios = require('axios');
 async function main(args) {
@@ -148,7 +156,7 @@ class CodeExec(ToolBase, ABC):
             self.set_output("_ERROR", "construct code request error: " + str(e))
 
         try:
-            resp = requests.post(url=f"http://{settings.SANDBOX_HOST}:9385/run", json=code_req, timeout=10)
+            resp = requests.post(url=f"http://{settings.SANDBOX_HOST}:9385/run", json=code_req, timeout=os.environ.get("COMPONENT_EXEC_TIMEOUT", 10*60))
             logging.info(f"http://{settings.SANDBOX_HOST}:9385/run", code_req, resp.status_code)
             if resp.status_code != 200:
                 resp.raise_for_status()

agent/tools/exesql.py

@@ -14,6 +14,7 @@
 #  limitations under the License.
 #
 import os
+import re
 from abc import ABC
 import pandas as pd
 import pymysql
@@ -78,6 +79,17 @@ class ExeSQL(ToolBase, ABC):
 
     @timeout(os.environ.get("COMPONENT_EXEC_TIMEOUT", 60))
     def _invoke(self, **kwargs):
+
+        def convert_decimals(obj):
+            from decimal import Decimal
+            if isinstance(obj, Decimal):
+                return float(obj)  # 或 str(obj)
+            elif isinstance(obj, dict):
+                return {k: convert_decimals(v) for k, v in obj.items()}
+            elif isinstance(obj, list):
+                return [convert_decimals(item) for item in obj]
+            return obj
+
         sql = kwargs.get("sql")
         if not sql:
             raise Exception("SQL for `ExeSQL` MUST not be empty.")
@@ -109,7 +121,7 @@ class ExeSQL(ToolBase, ABC):
             single_sql = single_sql.replace('```','')
             if not single_sql:
                 continue
-
+            single_sql = re.sub(r"\[ID:[0-9]+\]", "", single_sql)
             cursor.execute(single_sql)
             if cursor.rowcount == 0:
                 sql_res.append({"content": "No record in the database!"})
@@ -121,7 +133,11 @@ class ExeSQL(ToolBase, ABC):
                 single_res = pd.DataFrame([i for i in cursor.fetchmany(self._param.max_records)])
                 single_res.columns = [i[0] for i in cursor.description]
 
-            sql_res.append(single_res.to_dict(orient='records'))
+            for col in single_res.columns:
+                if pd.api.types.is_datetime64_any_dtype(single_res[col]):
+                    single_res[col] = single_res[col].dt.strftime('%Y-%m-%d')
+
+            sql_res.append(convert_decimals(single_res.to_dict(orient='records')))
             formalized_content.append(single_res.to_markdown(index=False, floatfmt=".6f"))
 
         self.set_output("json", sql_res)
@@ -129,4 +145,4 @@ class ExeSQL(ToolBase, ABC):
         return self.output("formalized_content")
 
     def thoughts(self) -> str:
-        return "Query sent—waiting for the data."
+        return "Query sent—waiting for the data."

agent/tools/retrieval.py

@@ -86,10 +86,16 @@ class Retrieval(ToolBase, ABC):
                 kb_ids.append(id)
                 continue
             kb_nm = self._canvas.get_variable_value(id)
-            e, kb = KnowledgebaseService.get_by_name(kb_nm)
-            if not e:
-                raise Exception(f"Dataset({kb_nm}) does not exist.")
-            kb_ids.append(kb.id)
+            # if kb_nm is a list
+            kb_nm_list = kb_nm if isinstance(kb_nm, list) else [kb_nm]
+            for nm_or_id in kb_nm_list:
+                e, kb = KnowledgebaseService.get_by_name(nm_or_id,
+                                                         self._canvas._tenant_id)
+                if not e:
+                    e, kb = KnowledgebaseService.get_by_id(nm_or_id)
+                    if not e:
+                        raise Exception(f"Dataset({nm_or_id}) does not exist.")
+                kb_ids.append(kb.id)
 
         filtered_kb_ids: list[str] = list(set([kb_id for kb_id in kb_ids if kb_id]))
 
@@ -108,7 +114,9 @@ class Retrieval(ToolBase, ABC):
         if self._param.rerank_id:
             rerank_mdl = LLMBundle(kbs[0].tenant_id, LLMType.RERANK, self._param.rerank_id)
 
-        query = kwargs["query"]
+        vars = self.get_input_elements_from_text(kwargs["query"])
+        vars = {k:o["value"] for k,o in vars.items()}
+        query = self.string_format(kwargs["query"], vars)
         if self._param.cross_languages:
             query = cross_languages(kbs[0].tenant_id, None, query, self._param.cross_languages)
 

agentic_reasoning/prompts.py

@@ -20,94 +20,128 @@ BEGIN_SEARCH_RESULT = "<|begin_search_result|>"
 END_SEARCH_RESULT = "<|end_search_result|>"
 MAX_SEARCH_LIMIT = 6
 
-REASON_PROMPT = (
-        "You are a reasoning assistant with the ability to perform dataset searches to help "
-        "you answer the user's question accurately. You have special tools:\n\n"
-        f"- To perform a search: write {BEGIN_SEARCH_QUERY} your query here {END_SEARCH_QUERY}.\n"
-        f"Then, the system will search and analyze relevant content, then provide you with helpful information in the format {BEGIN_SEARCH_RESULT} ...search results... {END_SEARCH_RESULT}.\n\n"
-        f"You can repeat the search process multiple times if necessary. The maximum number of search attempts is limited to {MAX_SEARCH_LIMIT}.\n\n"
-        "Once you have all the information you need, continue your reasoning.\n\n"
-        "-- Example 1 --\n" ########################################
-        "Question: \"Are both the directors of Jaws and Casino Royale from the same country?\"\n"
-        "Assistant:\n"
-        f"    {BEGIN_SEARCH_QUERY}Who is the director of Jaws?{END_SEARCH_QUERY}\n\n"
-        "User:\n"
-        f"    {BEGIN_SEARCH_RESULT}\nThe director of Jaws is Steven Spielberg...\n{END_SEARCH_RESULT}\n\n"
-        "Continues reasoning with the new information.\n"
-        "Assistant:\n"
-        f"    {BEGIN_SEARCH_QUERY}Where is Steven Spielberg from?{END_SEARCH_QUERY}\n\n"
-        "User:\n"
-        f"    {BEGIN_SEARCH_RESULT}\nSteven Allan Spielberg is an American filmmaker...\n{END_SEARCH_RESULT}\n\n"
-        "Continues reasoning with the new information...\n\n"
-        "Assistant:\n"
-        f"    {BEGIN_SEARCH_QUERY}Who is the director of Casino Royale?{END_SEARCH_QUERY}\n\n"
-        "User:\n"
-        f"    {BEGIN_SEARCH_RESULT}\nCasino Royale is a 2006 spy film directed by Martin Campbell...\n{END_SEARCH_RESULT}\n\n"
-        "Continues reasoning with the new information...\n\n"
-        "Assistant:\n"
-        f"    {BEGIN_SEARCH_QUERY}Where is Martin Campbell from?{END_SEARCH_QUERY}\n\n"
-        "User:\n"
-        f"    {BEGIN_SEARCH_RESULT}\nMartin Campbell (born 24 October 1943) is a New Zealand film and television director...\n{END_SEARCH_RESULT}\n\n"
-        "Continues reasoning with the new information...\n\n"
-        "Assistant:\nIt's enough to answer the question\n"
+REASON_PROMPT = f"""You are an advanced reasoning agent. Your goal is to answer the user's question by breaking it down into a series of verifiable steps.
 
-        "-- Example 2 --\n" #########################################
-        "Question: \"When was the founder of craigslist born?\"\n"
-        "Assistant:\n"
-        f"    {BEGIN_SEARCH_QUERY}Who was the founder of craigslist?{END_SEARCH_QUERY}\n\n"
-        "User:\n"
-        f"    {BEGIN_SEARCH_RESULT}\nCraigslist was founded by Craig Newmark...\n{END_SEARCH_RESULT}\n\n"
-        "Continues reasoning with the new information.\n"
-        "Assistant:\n"
-        f"    {BEGIN_SEARCH_QUERY} When was Craig Newmark born?{END_SEARCH_QUERY}\n\n"
-        "User:\n"
-        f"    {BEGIN_SEARCH_RESULT}\nCraig Newmark was born on December 6, 1952...\n{END_SEARCH_RESULT}\n\n"
-        "Continues reasoning with the new information...\n\n"
-        "Assistant:\nIt's enough to answer the question\n"
-        "**Remember**:\n"
-        f"- You have a dataset to search, so you just provide a proper search query.\n"
-        f"- Use {BEGIN_SEARCH_QUERY} to request a dataset search and end with {END_SEARCH_QUERY}.\n"
-        "- The language of query MUST be as the same as 'Question' or 'search result'.\n"
-        "- If no helpful information can be found, rewrite the search query to be less and precise keywords.\n"
-        "- When done searching, continue your reasoning.\n\n"
-        'Please answer the following question. You should think step by step to solve it.\n\n'
-    )
+You have access to a powerful search tool to find information.
 
-RELEVANT_EXTRACTION_PROMPT = """**Task Instruction:**
+**Your Task:**
+1.  Analyze the user's question.
+2.  If you need information, issue a search query to find a specific fact.
+3.  Review the search results.
+4.  Repeat the search process until you have all the facts needed to answer the question.
+5.  Once you have gathered sufficient information, synthesize the facts and provide the final answer directly.
 
-    You are tasked with reading and analyzing web pages based on the following inputs: **Previous Reasoning Steps**, **Current Search Query**, and **Searched Web Pages**. Your objective is to extract relevant and helpful information for **Current Search Query** from the **Searched Web Pages** and seamlessly integrate this information into the **Previous Reasoning Steps** to continue reasoning for the original question.
+**Tool Usage:**
+- To search, you MUST write your query between the special tokens: {BEGIN_SEARCH_QUERY}your query{END_SEARCH_QUERY}.
+- The system will provide results between {BEGIN_SEARCH_RESULT}search results{END_SEARCH_RESULT}.
+- You have a maximum of {MAX_SEARCH_LIMIT} search attempts.
 
-    **Guidelines:**
+---
+**Example 1: Multi-hop Question**
 
-    1. **Analyze the Searched Web Pages:**
-    - Carefully review the content of each searched web page.
-    - Identify factual information that is relevant to the **Current Search Query** and can aid in the reasoning process for the original question.
+**Question:** "Are both the directors of Jaws and Casino Royale from the same country?"
 
-    2. **Extract Relevant Information:**
-    - Select the information from the Searched Web Pages that directly contributes to advancing the **Previous Reasoning Steps**.
-    - Ensure that the extracted information is accurate and relevant.
+**Your Thought Process & Actions:**
+First, I need to identify the director of Jaws.
+{BEGIN_SEARCH_QUERY}who is the director of Jaws?{END_SEARCH_QUERY}
+[System returns search results]
+{BEGIN_SEARCH_RESULT}
+Jaws is a 1975 American thriller film directed by Steven Spielberg.
+{END_SEARCH_RESULT}
+Okay, the director of Jaws is Steven Spielberg. Now I need to find out his nationality.
+{BEGIN_SEARCH_QUERY}where is Steven Spielberg from?{END_SEARCH_QUERY}
+[System returns search results]
+{BEGIN_SEARCH_RESULT}
+Steven Allan Spielberg is an American filmmaker. Born in Cincinnati, Ohio...
+{END_SEARCH_RESULT}
+So, Steven Spielberg is from the USA. Next, I need to find the director of Casino Royale.
+{BEGIN_SEARCH_QUERY}who is the director of Casino Royale 2006?{END_SEARCH_QUERY}
+[System returns search results]
+{BEGIN_SEARCH_RESULT}
+Casino Royale is a 2006 spy film directed by Martin Campbell.
+{END_SEARCH_RESULT}
+The director of Casino Royale is Martin Campbell. Now I need his nationality.
+{BEGIN_SEARCH_QUERY}where is Martin Campbell from?{END_SEARCH_QUERY}
+[System returns search results]
+{BEGIN_SEARCH_RESULT}
+Martin Campbell (born 24 October 1943) is a New Zealand film and television director.
+{END_SEARCH_RESULT}
+I have all the information. Steven Spielberg is from the USA, and Martin Campbell is from New Zealand. They are not from the same country.
 
-    3. **Output Format:**
-    - **If the web pages provide helpful information for current search query:** Present the information beginning with `**Final Information**` as shown below.
-    - The language of query **MUST BE** as the same as 'Search Query' or 'Web Pages'.\n"
-    **Final Information**
+Final Answer: No, the directors of Jaws and Casino Royale are not from the same country. Steven Spielberg is from the USA, and Martin Campbell is from New Zealand.
 
-    [Helpful information]
+---
+**Example 2: Simple Fact Retrieval**
 
-    - **If the web pages do not provide any helpful information for current search query:** Output the following text.
+**Question:** "When was the founder of craigslist born?"
 
-    **Final Information**
+**Your Thought Process & Actions:**
+First, I need to know who founded craigslist.
+{BEGIN_SEARCH_QUERY}who founded craigslist?{END_SEARCH_QUERY}
+[System returns search results]
+{BEGIN_SEARCH_RESULT}
+Craigslist was founded in 1995 by Craig Newmark.
+{END_SEARCH_RESULT}
+The founder is Craig Newmark. Now I need his birth date.
+{BEGIN_SEARCH_QUERY}when was Craig Newmark born?{END_SEARCH_QUERY}
+[System returns search results]
+{BEGIN_SEARCH_RESULT}
+Craig Newmark was born on December 6, 1952.
+{END_SEARCH_RESULT}
+I have found the answer.
 
-    No helpful information found.
+Final Answer: The founder of craigslist, Craig Newmark, was born on December 6, 1952.
 
-    **Inputs:**
-    - **Previous Reasoning Steps:**  
-    {prev_reasoning}
+---
+**Important Rules:**
+- **One Fact at a Time:** Decompose the problem and issue one search query at a time to find a single, specific piece of information.
+- **Be Precise:** Formulate clear and precise search queries. If a search fails, rephrase it.
+- **Synthesize at the End:** Do not provide the final answer until you have completed all necessary searches.
+- **Language Consistency:** Your search queries should be in the same language as the user's question.
 
-    - **Current Search Query:**  
-    {search_query}
+Now, begin your work. Please answer the following question by thinking step-by-step.
+"""
 
-    - **Searched Web Pages:**  
-    {document}
+RELEVANT_EXTRACTION_PROMPT = """You are a highly efficient information extraction module. Your sole purpose is to extract the single most relevant piece of information from the provided `Searched Web Pages` that directly answers the `Current Search Query`.
 
-    """
+**Your Task:**
+1.  Read the `Current Search Query` to understand what specific information is needed.
+2.  Scan the `Searched Web Pages` to find the answer to that query.
+3.  Extract only the essential, factual information that answers the query. Be concise.
+
+**Context (For Your Information Only):**
+The `Previous Reasoning Steps` are provided to give you context on the overall goal, but your primary focus MUST be on answering the `Current Search Query`. Do not use information from the previous steps in your output.
+
+**Output Format:**
+Your response must follow one of two formats precisely.
+
+1.  **If a direct and relevant answer is found:**
+    - Start your response immediately with `Final Information`.
+    - Provide only the extracted fact(s). Do not add any extra conversational text.
+
+    *Example:*
+    `Current Search Query`: Where is Martin Campbell from?
+    `Searched Web Pages`: [Long article snippet about Martin Campbell's career, which includes the sentence "Martin Campbell (born 24 October 1943) is a New Zealand film and television director..."]
+    
+    *Your Output:*
+    Final Information
+    Martin Campbell is a New Zealand film and television director.
+
+2.  **If no relevant answer that directly addresses the query is found in the web pages:**
+    - Start your response immediately with `Final Information`.
+    - Write the exact phrase: `No helpful information found.`
+
+---
+**BEGIN TASK**
+
+**Inputs:**
+
+- **Previous Reasoning Steps:**
+{prev_reasoning}
+
+- **Current Search Query:**
+{search_query}
+
+- **Searched Web Pages:**
+{document}
+"""

api/apps/__init__.py

@@ -29,6 +29,7 @@ from api.db.db_models import close_connection
 from api.db.services import UserService
 from api.utils import CustomJSONEncoder, commands
 
+from flask_mail import Mail
 from flask_session import Session
 from flask_login import LoginManager
 from api import settings
@@ -40,6 +41,7 @@ __all__ = ["app"]
 Request.json = property(lambda self: self.get_json(force=True, silent=True))
 
 app = Flask(__name__)
+smtp_mail_server = Mail()
 
 # Add this at the beginning of your file to configure Swagger UI
 swagger_config = {
@@ -146,16 +148,16 @@ def load_user(web_request):
     if authorization:
         try:
             access_token = str(jwt.loads(authorization))
-            
+
             if not access_token or not access_token.strip():
                 logging.warning("Authentication attempt with empty access token")
                 return None
-            
+
             # Access tokens should be UUIDs (32 hex characters)
             if len(access_token.strip()) < 32:
                 logging.warning(f"Authentication attempt with invalid token format: {len(access_token)} chars")
                 return None
-            
+
             user = UserService.query(
                 access_token=access_token, status=StatusEnum.VALID.value
             )

api/apps/canvas_app.py

@@ -32,8 +32,7 @@ from api.db.services.user_service import TenantService
 from api.db.services.user_canvas_version import UserCanvasVersionService
 from api.settings import RetCode
 from api.utils import get_uuid
-from api.utils.api_utils import get_json_result, server_error_response, validate_request, get_data_error_result, \
-    get_error_data_result
+from api.utils.api_utils import get_json_result, server_error_response, validate_request, get_data_error_result
 from agent.canvas import Canvas
 from peewee import MySQLDatabase, PostgresqlDatabase
 from api.db.db_models import APIToken
@@ -62,7 +61,7 @@ def canvas_list():
 @login_required
 def rm():
     for i in request.json["canvas_ids"]:
-        if not UserCanvasService.query(user_id=current_user.id,id=i):
+        if not UserCanvasService.accessible(i, current_user.id):
             return get_json_result(
                 data=False, message='Only owner of canvas authorized for this operation.',
                 code=RetCode.OPERATING_ERROR)
@@ -75,23 +74,23 @@ def rm():
 @login_required
 def save():
     req = request.json
-    req["user_id"] = current_user.id
     if not isinstance(req["dsl"], str):
         req["dsl"] = json.dumps(req["dsl"], ensure_ascii=False)
     req["dsl"] = json.loads(req["dsl"])
     if "id" not in req:
+        req["user_id"] = current_user.id
         if UserCanvasService.query(user_id=current_user.id, title=req["title"].strip()):
             return get_data_error_result(message=f"{req['title'].strip()} already exists.")
         req["id"] = get_uuid()
         if not UserCanvasService.save(**req):
             return get_data_error_result(message="Fail to save canvas.")
     else:
-        if not UserCanvasService.query(user_id=current_user.id, id=req["id"]):
+        if not UserCanvasService.accessible(req["id"], current_user.id):
             return get_json_result(
                 data=False, message='Only owner of canvas authorized for this operation.',
                 code=RetCode.OPERATING_ERROR)
         UserCanvasService.update_by_id(req["id"], req)
-    # save version    
+    # save version
     UserCanvasVersionService.insert( user_canvas_id=req["id"], dsl=req["dsl"], title="{0}_{1}".format(req["title"], time.strftime("%Y_%m_%d_%H_%M_%S")))
     UserCanvasVersionService.delete_all_versions(req["id"])
     return get_json_result(data=req)
@@ -100,9 +99,9 @@ def save():
 @manager.route('/get/<canvas_id>', methods=['GET'])  # noqa: F821
 @login_required
 def get(canvas_id):
-    e, c = UserCanvasService.get_by_tenant_id(canvas_id)
-    if not e or c["user_id"] != current_user.id:
+    if not UserCanvasService.accessible(canvas_id, current_user.id):
         return get_data_error_result(message="canvas not found.")
+    e, c = UserCanvasService.get_by_tenant_id(canvas_id)
     return get_json_result(data=c)
 
 
@@ -116,6 +115,12 @@ def getsse(canvas_id):
     if not objs:
         return get_data_error_result(message='Authentication error: API key is invalid!"')
     tenant_id = objs[0].tenant_id
+    if not UserCanvasService.query(user_id=tenant_id, id=canvas_id):
+        return get_json_result(
+            data=False,
+            message='Only owner of canvas authorized for this operation.',
+            code=RetCode.OPERATING_ERROR
+        )
     e, c = UserCanvasService.get_by_id(canvas_id)
     if not e or c.user_id != tenant_id:
         return get_data_error_result(message="canvas not found.")
@@ -131,14 +136,15 @@ def run():
     files = req.get("files", [])
     inputs = req.get("inputs", {})
     user_id = req.get("user_id", current_user.id)
-    e, cvs = UserCanvasService.get_by_id(req["id"])
-    if not e:
-        return get_data_error_result(message="canvas not found.")
-    if not UserCanvasService.query(user_id=current_user.id, id=req["id"]):
+    if not UserCanvasService.accessible(req["id"], current_user.id):
         return get_json_result(
             data=False, message='Only owner of canvas authorized for this operation.',
             code=RetCode.OPERATING_ERROR)
 
+    e, cvs = UserCanvasService.get_by_id(req["id"])
+    if not e:
+        return get_data_error_result(message="canvas not found.")
+
     if not isinstance(cvs.dsl, str):
         cvs.dsl = json.dumps(cvs.dsl, ensure_ascii=False)
 
@@ -172,14 +178,14 @@ def run():
 @login_required
 def reset():
     req = request.json
+    if not UserCanvasService.accessible(req["id"], current_user.id):
+        return get_json_result(
+            data=False, message='Only owner of canvas authorized for this operation.',
+            code=RetCode.OPERATING_ERROR)
     try:
         e, user_canvas = UserCanvasService.get_by_id(req["id"])
         if not e:
             return get_data_error_result(message="canvas not found.")
-        if not UserCanvasService.query(user_id=current_user.id, id=req["id"]):
-            return get_json_result(
-                data=False, message='Only owner of canvas authorized for this operation.',
-                code=RetCode.OPERATING_ERROR)
 
         canvas = Canvas(json.dumps(user_canvas.dsl), current_user.id)
         canvas.reset()
@@ -290,15 +296,12 @@ def input_form():
 @login_required
 def debug():
     req = request.json
+    if not UserCanvasService.accessible(req["id"], current_user.id):
+        return get_json_result(
+            data=False, message='Only owner of canvas authorized for this operation.',
+            code=RetCode.OPERATING_ERROR)
     try:
         e, user_canvas = UserCanvasService.get_by_id(req["id"])
-        if not e:
-            return get_data_error_result(message="canvas not found.")
-        if not UserCanvasService.query(user_id=current_user.id, id=req["id"]):
-            return get_json_result(
-                data=False, message='Only owner of canvas authorized for this operation.',
-                code=RetCode.OPERATING_ERROR)
-
         canvas = Canvas(json.dumps(user_canvas.dsl), current_user.id)
         canvas.reset()
         canvas.message_id = get_uuid()
@@ -350,7 +353,7 @@ def test_db_connect():
         if req["db_type"] != 'mssql':
             db.connect()
         db.close()
-        
+
         return get_json_result(data="Database Connection Successful!")
     except Exception as e:
         return server_error_response(e)
@@ -372,7 +375,7 @@ def getlistversion(canvas_id):
 @login_required
 def getversion( version_id):
     try:
-      
+
         e, version = UserCanvasVersionService.get_by_id(version_id)
         if version:
             return get_json_result(data=version.to_dict())
@@ -382,7 +385,7 @@ def getversion( version_id):
 
 @manager.route('/listteam', methods=['GET'])  # noqa: F821
 @login_required
-def list_kbs():
+def list_canvas():
     keywords = request.args.get("keywords", "")
     page_number = int(request.args.get("page", 1))
     items_per_page = int(request.args.get("page_size", 150))
@@ -390,10 +393,10 @@ def list_kbs():
     desc = request.args.get("desc", True)
     try:
         tenants = TenantService.get_joined_tenants_by_user_id(current_user.id)
-        kbs, total = UserCanvasService.get_by_tenant_ids(
+        canvas, total = UserCanvasService.get_by_tenant_ids(
             [m["tenant_id"] for m in tenants], current_user.id, page_number,
             items_per_page, orderby, desc, keywords)
-        return get_json_result(data={"kbs": kbs, "total": total})
+        return get_json_result(data={"canvas": canvas, "total": total})
     except Exception as e:
         return server_error_response(e)
 
@@ -404,6 +407,12 @@ def list_kbs():
 def setting():
     req = request.json
     req["user_id"] = current_user.id
+
+    if not UserCanvasService.accessible(req["id"], current_user.id):
+        return get_json_result(
+            data=False, message='Only owner of canvas authorized for this operation.',
+            code=RetCode.OPERATING_ERROR)
+
     e,flow = UserCanvasService.get_by_id(req["id"])
     if not e:
         return get_data_error_result(message="canvas not found.")
@@ -415,10 +424,7 @@ def setting():
         flow["permission"] = req["permission"]
     if req["avatar"]:
         flow["avatar"] = req["avatar"]
-    if not UserCanvasService.query(user_id=current_user.id, id=req["id"]):
-        return get_json_result(
-            data=False, message='Only owner of canvas authorized for this operation.',
-            code=RetCode.OPERATING_ERROR)
+
     num= UserCanvasService.update_by_id(req["id"], flow)
     return get_json_result(data=num)
 
@@ -441,8 +447,10 @@ def trace():
 @login_required
 def sessions(canvas_id):
     tenant_id = current_user.id
-    if not UserCanvasService.query(user_id=tenant_id, id=canvas_id):
-        return get_error_data_result(message=f"You don't own the agent {canvas_id}.")
+    if not UserCanvasService.accessible(canvas_id, tenant_id):
+        return get_json_result(
+            data=False, message='Only owner of canvas authorized for this operation.',
+            code=RetCode.OPERATING_ERROR)
 
     user_id = request.args.get("user_id")
     page_number = int(request.args.get("page", 1))

api/apps/chunk_app.py

@@ -23,15 +23,18 @@ from flask_login import current_user, login_required
 
 from api import settings
 from api.db import LLMType, ParserType
+from api.db.services.dialog_service import meta_filter
 from api.db.services.document_service import DocumentService
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.llm_service import LLMBundle
+from api.db.services.search_service import SearchService
 from api.db.services.user_service import UserTenantService
 from api.utils.api_utils import get_data_error_result, get_json_result, server_error_response, validate_request
 from rag.app.qa import beAdoc, rmPrefix
 from rag.app.tag import label_question
 from rag.nlp import rag_tokenizer, search
 from rag.prompts import cross_languages, keyword_extraction
+from rag.prompts.prompts import gen_meta_filter
 from rag.settings import PAGERANK_FLD
 from rag.utils import rmSpace
 
@@ -288,13 +291,26 @@ def retrieval_test():
     if isinstance(kb_ids, str):
         kb_ids = [kb_ids]
     doc_ids = req.get("doc_ids", [])
-    similarity_threshold = float(req.get("similarity_threshold", 0.0))
-    vector_similarity_weight = float(req.get("vector_similarity_weight", 0.3))
     use_kg = req.get("use_kg", False)
     top = int(req.get("top_k", 1024))
     langs = req.get("cross_languages", [])
     tenant_ids = []
 
+    if req.get("search_id", ""):
+        search_config = SearchService.get_detail(req.get("search_id", "")).get("search_config", {})
+        meta_data_filter = search_config.get("meta_data_filter", {})
+        metas = DocumentService.get_meta_by_kbs(kb_ids)
+        if meta_data_filter.get("method") == "auto":
+            chat_mdl = LLMBundle(current_user.id, LLMType.CHAT, llm_name=search_config.get("chat_id", ""))
+            filters = gen_meta_filter(chat_mdl, metas, question)
+            doc_ids.extend(meta_filter(metas, filters))
+            if not doc_ids:
+                doc_ids = None
+        elif meta_data_filter.get("method") == "manual":
+            doc_ids.extend(meta_filter(metas, meta_data_filter["manual"]))
+            if not doc_ids:
+                doc_ids = None
+
     try:
         tenants = UserTenantService.query(user_id=current_user.id)
         for kb_id in kb_ids:
@@ -327,7 +343,9 @@ def retrieval_test():
 
         labels = label_question(question, [kb])
         ranks = settings.retrievaler.retrieval(question, embd_mdl, tenant_ids, kb_ids, page, size,
-                               similarity_threshold, vector_similarity_weight, top,
+                               float(req.get("similarity_threshold", 0.0)),
+                               float(req.get("vector_similarity_weight", 0.3)),
+                               top,
                                doc_ids, rerank_mdl=rerank_mdl, highlight=req.get("highlight"),
                                rank_feature=labels
                                )

api/apps/conversation_app.py

@@ -17,22 +17,19 @@ import json
 import re
 import traceback
 from copy import deepcopy
-
-import trio
 from flask import Response, request
 from flask_login import current_user, login_required
-
 from api import settings
 from api.db import LLMType
 from api.db.db_models import APIToken
 from api.db.services.conversation_service import ConversationService, structure_answer
-from api.db.services.dialog_service import DialogService, ask, chat
-from api.db.services.knowledgebase_service import KnowledgebaseService
-from api.db.services.llm_service import LLMBundle, TenantService
-from api.db.services.user_service import UserTenantService
+from api.db.services.dialog_service import DialogService, ask, chat, gen_mindmap
+from api.db.services.llm_service import LLMBundle
+from api.db.services.search_service import SearchService
+from api.db.services.tenant_llm_service import TenantLLMService
+from api.db.services.user_service import TenantService, UserTenantService
 from api.utils.api_utils import get_data_error_result, get_json_result, server_error_response, validate_request
-from graphrag.general.mind_map_extractor import MindMapExtractor
-from rag.app.tag import label_question
+from rag.prompts.prompt_template import load_prompt
 from rag.prompts.prompts import chunks_format
 
 
@@ -66,7 +63,14 @@ def set_conversation():
         e, dia = DialogService.get_by_id(req["dialog_id"])
         if not e:
             return get_data_error_result(message="Dialog not found")
-        conv = {"id": conv_id, "dialog_id": req["dialog_id"], "name": name, "message": [{"role": "assistant", "content": dia.prompt_config["prologue"]}],"user_id": current_user.id}
+        conv = {
+            "id": conv_id,
+            "dialog_id": req["dialog_id"],
+            "name": name,
+            "message": [{"role": "assistant", "content": dia.prompt_config["prologue"]}],
+            "user_id": current_user.id,
+            "reference": [],
+        }
         ConversationService.save(**conv)
         return get_json_result(data=conv)
     except Exception as e:
@@ -173,6 +177,21 @@ def completion():
             continue
         msg.append(m)
     message_id = msg[-1].get("id")
+    chat_model_id = req.get("llm_id", "")
+    req.pop("llm_id", None)
+
+    chat_model_config = {}
+    for model_config in [
+        "temperature",
+        "top_p",
+        "frequency_penalty",
+        "presence_penalty",
+        "max_tokens",
+    ]:
+        config = req.get(model_config)
+        if config:
+            chat_model_config[model_config] = config
+
     try:
         e, conv = ConversationService.get_by_id(req["conversation_id"])
         if not e:
@@ -186,23 +205,26 @@ def completion():
 
         if not conv.reference:
             conv.reference = []
-        else:
-            for ref in conv.reference:
-                if isinstance(ref, list):
-                    continue
-                ref["chunks"] = chunks_format(ref)
-
-        if not conv.reference:
-            conv.reference = []
+        conv.reference = [r for r in conv.reference if r]
         conv.reference.append({"chunks": [], "doc_aggs": []})
 
+        if chat_model_id:
+            if not TenantLLMService.get_api_key(tenant_id=dia.tenant_id, model_name=chat_model_id):
+                req.pop("chat_model_id", None)
+                req.pop("chat_model_config", None)
+                return get_data_error_result(message=f"Cannot use specified model {chat_model_id}.")
+            dia.llm_id = chat_model_id
+            dia.llm_setting = chat_model_config
+
+        is_embedded = bool(chat_model_id)
         def stream():
             nonlocal dia, msg, req, conv
             try:
                 for ans in chat(dia, msg, True, **req):
                     ans = structure_answer(conv, ans, message_id, conv.id)
                     yield "data:" + json.dumps({"code": 0, "message": "", "data": ans}, ensure_ascii=False) + "\n\n"
-                ConversationService.update_by_id(conv.id, conv.to_dict())
+                if not is_embedded:
+                    ConversationService.update_by_id(conv.id, conv.to_dict())
             except Exception as e:
                 traceback.print_exc()
                 yield "data:" + json.dumps({"code": 500, "message": str(e), "data": {"answer": "**ERROR**: " + str(e), "reference": []}}, ensure_ascii=False) + "\n\n"
@@ -220,7 +242,8 @@ def completion():
             answer = None
             for ans in chat(dia, msg, **req):
                 answer = structure_answer(conv, ans, message_id, conv.id)
-                ConversationService.update_by_id(conv.id, conv.to_dict())
+                if not is_embedded:
+                    ConversationService.update_by_id(conv.id, conv.to_dict())
                 break
             return get_json_result(data=answer)
     except Exception as e:
@@ -316,10 +339,18 @@ def ask_about():
     req = request.json
     uid = current_user.id
 
+    search_id = req.get("search_id", "")
+    search_app = None
+    search_config = {}
+    if search_id:
+        search_app = SearchService.get_detail(search_id)
+    if search_app:
+        search_config = search_app.get("search_config", {})
+
     def stream():
         nonlocal req, uid
         try:
-            for ans in ask(req["question"], req["kb_ids"], uid):
+            for ans in ask(req["question"], req["kb_ids"], uid, search_config=search_config):
                 yield "data:" + json.dumps({"code": 0, "message": "", "data": ans}, ensure_ascii=False) + "\n\n"
         except Exception as e:
             yield "data:" + json.dumps({"code": 500, "message": str(e), "data": {"answer": "**ERROR**: " + str(e), "reference": []}}, ensure_ascii=False) + "\n\n"
@@ -338,18 +369,14 @@ def ask_about():
 @validate_request("question", "kb_ids")
 def mindmap():
     req = request.json
-    kb_ids = req["kb_ids"]
-    e, kb = KnowledgebaseService.get_by_id(kb_ids[0])
-    if not e:
-        return get_data_error_result(message="Knowledgebase not found!")
+    search_id = req.get("search_id", "")
+    search_app = SearchService.get_detail(search_id) if search_id else {}
+    search_config = search_app.get("search_config", {}) if search_app else {}
+    kb_ids = search_config.get("kb_ids", [])
+    kb_ids.extend(req["kb_ids"])
+    kb_ids = list(set(kb_ids))
 
-    embd_mdl = LLMBundle(kb.tenant_id, LLMType.EMBEDDING, llm_name=kb.embd_id)
-    chat_mdl = LLMBundle(current_user.id, LLMType.CHAT)
-    question = req["question"]
-    ranks = settings.retrievaler.retrieval(question, embd_mdl, kb.tenant_id, kb_ids, 1, 12, 0.3, 0.3, aggs=False, rank_feature=label_question(question, [kb]))
-    mindmap = MindMapExtractor(chat_mdl)
-    mind_map = trio.run(mindmap, [c["content_with_weight"] for c in ranks["chunks"]])
-    mind_map = mind_map.output
+    mind_map = gen_mindmap(req["question"], kb_ids, search_app.get("tenant_id", current_user.id), search_config)
     if "error" in mind_map:
         return server_error_response(Exception(mind_map["error"]))
     return get_json_result(data=mind_map)
@@ -360,41 +387,20 @@ def mindmap():
 @validate_request("question")
 def related_questions():
     req = request.json
+
+    search_id = req.get("search_id", "")
+    search_config = {}
+    if search_id:
+        if search_app := SearchService.get_detail(search_id):
+            search_config = search_app.get("search_config", {})
+
     question = req["question"]
-    chat_mdl = LLMBundle(current_user.id, LLMType.CHAT)
-    prompt = """
-Role: You are an AI language model assistant tasked with generating 5-10 related questions based on a user’s original query. These questions should help expand the search query scope and improve search relevance.
 
-Instructions:
-	Input: You are provided with a user’s question.
-	Output: Generate 5-10 alternative questions that are related to the original user question. These alternatives should help retrieve a broader range of relevant documents from a vector database.
-	Context: Focus on rephrasing the original question in different ways, making sure the alternative questions are diverse but still connected to the topic of the original query. Do not create overly obscure, irrelevant, or unrelated questions.
-	Fallback: If you cannot generate any relevant alternatives, do not return any questions.
-	Guidance:
-	1. Each alternative should be unique but still relevant to the original query.
-	2. Keep the phrasing clear, concise, and easy to understand.
-	3. Avoid overly technical jargon or specialized terms unless directly relevant.
-	4. Ensure that each question contributes towards improving search results by broadening the search angle, not narrowing it.
+    chat_id = search_config.get("chat_id", "")
+    chat_mdl = LLMBundle(current_user.id, LLMType.CHAT, chat_id)
 
-Example:
-Original Question: What are the benefits of electric vehicles?
-
-Alternative Questions:
-	1. How do electric vehicles impact the environment?
-	2. What are the advantages of owning an electric car?
-	3. What is the cost-effectiveness of electric vehicles?
-	4. How do electric vehicles compare to traditional cars in terms of fuel efficiency?
-	5. What are the environmental benefits of switching to electric cars?
-	6. How do electric vehicles help reduce carbon emissions?
-	7. Why are electric vehicles becoming more popular?
-	8. What are the long-term savings of using electric vehicles?
-	9. How do electric vehicles contribute to sustainability?
-	10. What are the key benefits of electric vehicles for consumers?
-
-Reason:
-	Rephrasing the original query into multiple alternative questions helps the user explore different aspects of their search topic, improving the quality of search results.
-	These questions guide the search engine to provide a more comprehensive set of relevant documents.
-"""
+    gen_conf = search_config.get("llm_setting", {"temperature": 0.9})
+    prompt = load_prompt("related_question")
     ans = chat_mdl.chat(
         prompt,
         [
@@ -406,6 +412,6 @@ Related search terms:
     """,
             }
         ],
-        {"temperature": 0.9},
+        gen_conf,
     )
     return get_json_result(data=[re.sub(r"^[0-9]\. ", "", a) for a in ans.split("\n") if re.match(r"^[0-9]\. ", a)])

api/apps/dialog_app.py

@@ -16,9 +16,10 @@
 
 from flask import request
 from flask_login import login_required, current_user
+from api.db.services import duplicate_name
 from api.db.services.dialog_service import DialogService
 from api.db import StatusEnum
-from api.db.services.llm_service import TenantLLMService
+from api.db.services.tenant_llm_service import TenantLLMService
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.user_service import TenantService, UserTenantService
 from api import settings
@@ -32,7 +33,8 @@ from api.utils.api_utils import get_json_result
 @login_required
 def set_dialog():
     req = request.json
-    dialog_id = req.get("dialog_id")
+    dialog_id = req.get("dialog_id", "")
+    is_create = not dialog_id
     name = req.get("name", "New Dialog")
     if not isinstance(name, str):
         return get_data_error_result(message="Dialog name must be string.")
@@ -40,6 +42,15 @@ def set_dialog():
         return get_data_error_result(message="Dialog name can't be empty.")
     if len(name.encode("utf-8")) > 255:
         return get_data_error_result(message=f"Dialog name length is {len(name)} which is larger than 255")
+
+    if is_create and DialogService.query(tenant_id=current_user.id, name=name.strip()):
+        name = name.strip()
+        name = duplicate_name(
+            DialogService.query,
+            name=name,
+            tenant_id=current_user.id,
+            status=StatusEnum.VALID.value)
+
     description = req.get("description", "A helpful dialog")
     icon = req.get("icon", "")
     top_n = req.get("top_n", 6)
@@ -50,17 +61,19 @@ def set_dialog():
     similarity_threshold = req.get("similarity_threshold", 0.1)
     vector_similarity_weight = req.get("vector_similarity_weight", 0.3)
     llm_setting = req.get("llm_setting", {})
+    meta_data_filter = req.get("meta_data_filter", {})
     prompt_config = req["prompt_config"]
 
-    if not req.get("kb_ids", []) and not prompt_config.get("tavily_api_key") and "{knowledge}" in prompt_config['system']:
-        return get_data_error_result(message="Please remove `{knowledge}` in system prompt since no knowledge base/Tavily used here.")
+    if not is_create:
+        if not req.get("kb_ids", []) and not prompt_config.get("tavily_api_key") and "{knowledge}" in prompt_config['system']:
+            return get_data_error_result(message="Please remove `{knowledge}` in system prompt since no knowledge base/Tavily used here.")
 
-    for p in prompt_config["parameters"]:
-        if p["optional"]:
-            continue
-        if prompt_config["system"].find("{%s}" % p["key"]) < 0:
-            return get_data_error_result(
-                message="Parameter '{}' is not used".format(p["key"]))
+        for p in prompt_config["parameters"]:
+            if p["optional"]:
+                continue
+            if prompt_config["system"].find("{%s}" % p["key"]) < 0:
+                return get_data_error_result(
+                    message="Parameter '{}' is not used".format(p["key"]))
 
     try:
         e, tenant = TenantService.get_by_id(current_user.id)
@@ -83,6 +96,7 @@ def set_dialog():
                 "llm_id": llm_id,
                 "llm_setting": llm_setting,
                 "prompt_config": prompt_config,
+                "meta_data_filter": meta_data_filter,
                 "top_n": top_n,
                 "top_k": top_k,
                 "rerank_id": rerank_id,
@@ -153,6 +167,43 @@ def list_dialogs():
         return server_error_response(e)
 
 
+@manager.route('/next', methods=['POST'])  # noqa: F821
+@login_required
+def list_dialogs_next():
+    keywords = request.args.get("keywords", "")
+    page_number = int(request.args.get("page", 0))
+    items_per_page = int(request.args.get("page_size", 0))
+    parser_id = request.args.get("parser_id")
+    orderby = request.args.get("orderby", "create_time")
+    if request.args.get("desc", "true").lower() == "false":
+        desc = False
+    else:
+        desc = True
+
+    req = request.get_json()
+    owner_ids = req.get("owner_ids", [])
+    try:
+        if not owner_ids:
+            # tenants = TenantService.get_joined_tenants_by_user_id(current_user.id)
+            # tenants = [tenant["tenant_id"] for tenant in tenants]
+            tenants = [] # keep it here
+            dialogs, total = DialogService.get_by_tenant_ids(
+                tenants, current_user.id, page_number,
+                items_per_page, orderby, desc, keywords, parser_id)
+        else:
+            tenants = owner_ids
+            dialogs, total = DialogService.get_by_tenant_ids(
+                tenants, current_user.id, 0,
+                0, orderby, desc, keywords, parser_id)
+            dialogs = [dialog for dialog in dialogs if dialog["tenant_id"] in tenants]
+            total = len(dialogs)
+            if page_number and items_per_page:
+                dialogs = dialogs[(page_number-1)*items_per_page:page_number*items_per_page]
+        return get_json_result(data={"dialogs": dialogs, "total": total})
+    except Exception as e:
+        return server_error_response(e)
+
+
 @manager.route('/rm', methods=['POST'])  # noqa: F821
 @login_required
 @validate_request("dialog_ids")

api/apps/document_app.py

@@ -166,6 +166,17 @@ def create():
         if DocumentService.query(name=req["name"], kb_id=kb_id):
             return get_data_error_result(message="Duplicated document name in the same knowledgebase.")
 
+        kb_root_folder = FileService.get_kb_folder(kb.tenant_id)
+        if not kb_root_folder:
+            return get_data_error_result(message="Cannot find the root folder.")
+        kb_folder = FileService.new_a_file_from_kb(
+            kb.tenant_id,
+            kb.name,
+            kb_root_folder["id"],
+        )
+        if not kb_folder:
+            return get_data_error_result(message="Cannot find the kb folder for this file.")
+
         doc = DocumentService.insert(
             {
                 "id": get_uuid(),
@@ -180,6 +191,9 @@ def create():
                 "size": 0,
             }
         )
+
+        FileService.add_file_from_kb(doc.to_dict(), kb_folder["id"], kb.tenant_id)
+
         return get_json_result(data=doc.to_json())
     except Exception as e:
         return server_error_response(e)
@@ -206,6 +220,8 @@ def list_docs():
         desc = False
     else:
         desc = True
+    create_time_from = int(request.args.get("create_time_from", 0))
+    create_time_to = int(request.args.get("create_time_to", 0))
 
     req = request.get_json()
 
@@ -226,6 +242,14 @@ def list_docs():
     try:
         docs, tol = DocumentService.get_by_kb_id(kb_id, page_number, items_per_page, orderby, desc, keywords, run_status, types, suffix)
 
+        if create_time_from or create_time_to:
+            filtered_docs = []
+            for doc in docs:
+                doc_create_time = doc.get("create_time", 0)
+                if (create_time_from == 0 or doc_create_time >= create_time_from) and (create_time_to == 0 or doc_create_time <= create_time_to):
+                    filtered_docs.append(doc)
+            docs = filtered_docs
+
         for doc_item in docs:
             if doc_item["thumbnail"] and not doc_item["thumbnail"].startswith(IMG_BASE64_PREFIX):
                 doc_item["thumbnail"] = f"/v1/document/image/{kb_id}-{doc_item['thumbnail']}"
@@ -657,6 +681,11 @@ def set_meta():
         return get_json_result(data=False, message="No authorization.", code=settings.RetCode.AUTHENTICATION_ERROR)
     try:
         meta = json.loads(req["meta"])
+        if not isinstance(meta, dict):
+            return get_json_result(data=False, message="Only dictionary type supported.", code=settings.RetCode.ARGUMENT_ERROR)
+        for k,v in meta.items():
+            if not isinstance(v, str) and not isinstance(v, int) and not isinstance(v, float):
+                return get_json_result(data=False, message=f"The type is not supported: {v}", code=settings.RetCode.ARGUMENT_ERROR)
     except Exception as e:
         return get_json_result(data=False, message=f"Json syntax error: {e}", code=settings.RetCode.ARGUMENT_ERROR)
     if not isinstance(meta, dict):

api/apps/kb_app.py

@@ -247,7 +247,10 @@ def list_tags(kb_id):
             code=settings.RetCode.AUTHENTICATION_ERROR
         )
 
-    tags = settings.retrievaler.all_tags(current_user.id, [kb_id])
+    tenants = UserTenantService.get_tenants_by_user_id(current_user.id)
+    tags = []
+    for tenant in tenants:
+        tags += settings.retrievaler.all_tags(tenant["tenant_id"], [kb_id])
     return get_json_result(data=tags)
 
 
@@ -263,7 +266,10 @@ def list_tags_from_kbs():
                 code=settings.RetCode.AUTHENTICATION_ERROR
             )
 
-    tags = settings.retrievaler.all_tags(current_user.id, kb_ids)
+    tenants = UserTenantService.get_tenants_by_user_id(current_user.id)
+    tags = []
+    for tenant in tenants:
+        tags += settings.retrievaler.all_tags(tenant["tenant_id"], kb_ids)
     return get_json_result(data=tags)
 
 
@@ -345,6 +351,7 @@ def knowledge_graph(kb_id):
             obj["graph"]["edges"] = sorted(filtered_edges, key=lambda x: x.get("weight", 0), reverse=True)[:128]
     return get_json_result(data=obj)
 
+
 @manager.route('/<kb_id>/knowledge_graph', methods=['DELETE'])  # noqa: F821
 @login_required
 def delete_knowledge_graph(kb_id):
@@ -358,3 +365,17 @@ def delete_knowledge_graph(kb_id):
     settings.docStoreConn.delete({"knowledge_graph_kwd": ["graph", "subgraph", "entity", "relation"]}, search.index_name(kb.tenant_id), kb_id)
 
     return get_json_result(data=True)
+
+
+@manager.route("/get_meta", methods=["GET"])  # noqa: F821
+@login_required
+def get_meta():
+    kb_ids = request.args.get("kb_ids", "").split(",")
+    for kb_id in kb_ids:
+        if not KnowledgebaseService.accessible(kb_id, current_user.id):
+            return get_json_result(
+                data=False,
+                message='No authorization.',
+                code=settings.RetCode.AUTHENTICATION_ERROR
+            )
+    return get_json_result(data=DocumentService.get_meta_by_kbs(kb_ids))

api/apps/llm_app.py

@@ -15,16 +15,16 @@
 #
 import logging
 import json
-import base64
 from flask import request
 from flask_login import login_required, current_user
-from api.db.services.llm_service import LLMFactoriesService, TenantLLMService, LLMService
+from api.db.services.tenant_llm_service import LLMFactoriesService, TenantLLMService
+from api.db.services.llm_service import LLMService
 from api import settings
 from api.utils.api_utils import server_error_response, get_data_error_result, validate_request
 from api.db import StatusEnum, LLMType
 from api.db.db_models import TenantLLM
 from api.utils.api_utils import get_json_result
-from api.utils.base64_image import test_image_base64
+from api.utils.base64_image import test_image
 from rag.llm import EmbeddingModel, ChatModel, RerankModel, CvModel, TTSModel
 
 
@@ -58,6 +58,7 @@ def set_api_key():
     # test if api key works
     chat_passed, embd_passed, rerank_passed = False, False, False
     factory = req["llm_factory"]
+    extra = {"provider": factory}
     msg = ""
     for llm in LLMService.query(fid=factory):
         if not embd_passed and llm.model_type == LLMType.EMBEDDING.value:
@@ -74,7 +75,7 @@ def set_api_key():
         elif not chat_passed and llm.model_type == LLMType.CHAT.value:
             assert factory in ChatModel, f"Chat model from {factory} is not supported yet."
             mdl = ChatModel[factory](
-                req["api_key"], llm.llm_name, base_url=req.get("base_url"))
+                req["api_key"], llm.llm_name, base_url=req.get("base_url"), **extra)
             try:
                 m, tc = mdl.chat(None, [{"role": "user", "content": "Hello! How are you doing!"}],
                                  {"temperature": 0.9, 'max_tokens': 50})
@@ -82,7 +83,7 @@ def set_api_key():
                     raise Exception(m)
                 chat_passed = True
             except Exception as e:
-                msg += f"\nFail to access model({llm.llm_name}) using this api key." + str(
+                msg += f"\nFail to access model({llm.fid}/{llm.llm_name}) using this api key." + str(
                     e)
         elif not rerank_passed and llm.model_type == LLMType.RERANK:
             assert factory in RerankModel, f"Re-rank model from {factory} is not supported yet."
@@ -95,7 +96,7 @@ def set_api_key():
                 rerank_passed = True
                 logging.debug(f'passed model rerank {llm.llm_name}')
             except Exception as e:
-                msg += f"\nFail to access model({llm.llm_name}) using this api key." + str(
+                msg += f"\nFail to access model({llm.fid}/{llm.llm_name}) using this api key." + str(
                     e)
         if any([embd_passed, chat_passed, rerank_passed]):
             msg = ''
@@ -205,6 +206,7 @@ def add_llm():
 
     msg = ""
     mdl_nm = llm["llm_name"].split("___")[0]
+    extra = {"provider": factory}
     if llm["model_type"] == LLMType.EMBEDDING.value:
         assert factory in EmbeddingModel, f"Embedding model from {factory} is not supported yet."
         mdl = EmbeddingModel[factory](
@@ -222,7 +224,8 @@ def add_llm():
         mdl = ChatModel[factory](
             key=llm['api_key'],
             model_name=mdl_nm,
-            base_url=llm["api_base"]
+            base_url=llm["api_base"],
+            **extra,
         )
         try:
             m, tc = mdl.chat(None, [{"role": "user", "content": "Hello! How are you doing!"}], {
@@ -230,7 +233,7 @@ def add_llm():
             if not tc and m.find("**ERROR**:") >= 0:
                 raise Exception(m)
         except Exception as e:
-            msg += f"\nFail to access model({mdl_nm})." + str(
+            msg += f"\nFail to access model({factory}/{mdl_nm})." + str(
                 e)
     elif llm["model_type"] == LLMType.RERANK:
         assert factory in RerankModel, f"RE-rank model from {factory} is not supported yet."
@@ -244,9 +247,9 @@ def add_llm():
             if len(arr) == 0:
                 raise Exception("Not known.")
         except KeyError:
-            msg += f"{factory} dose not support this model({mdl_nm})"
+            msg += f"{factory} dose not support this model({factory}/{mdl_nm})"
         except Exception as e:
-            msg += f"\nFail to access model({mdl_nm})." + str(
+            msg += f"\nFail to access model({factory}/{mdl_nm})." + str(
                 e)
     elif llm["model_type"] == LLMType.IMAGE2TEXT.value:
         assert factory in CvModel, f"Image to text model from {factory} is not supported yet."
@@ -256,12 +259,12 @@ def add_llm():
             base_url=llm["api_base"]
         )
         try:
-            image_data = base64.b64decode(test_image_base64)
+            image_data = test_image
             m, tc = mdl.describe(image_data)
             if not m and not tc:
                 raise Exception(m)
         except Exception as e:
-            msg += f"\nFail to access model({mdl_nm})." + str(e)
+            msg += f"\nFail to access model({factory}/{mdl_nm})." + str(e)
     elif llm["model_type"] == LLMType.TTS:
         assert factory in TTSModel, f"TTS model from {factory} is not supported yet."
         mdl = TTSModel[factory](
@@ -271,7 +274,7 @@ def add_llm():
             for resp in mdl.tts("Hello~ Ragflower!"):
                 pass
         except RuntimeError as e:
-            msg += f"\nFail to access model({mdl_nm})." + str(e)
+            msg += f"\nFail to access model({factory}/{mdl_nm})." + str(e)
     else:
         # TODO: check other type of models
         pass
@@ -313,12 +316,12 @@ def delete_factory():
 def my_llms():
     try:
         include_details = request.args.get('include_details', 'false').lower() == 'true'
-        
+
         if include_details:
             res = {}
             objs = TenantLLMService.query(tenant_id=current_user.id)
             factories = LLMFactoriesService.query(status=StatusEnum.VALID.value)
-            
+
             for o in objs:
                 o_dict = o.to_dict()
                 factory_tags = None
@@ -326,13 +329,13 @@ def my_llms():
                     if f.name == o_dict["llm_factory"]:
                         factory_tags = f.tags
                         break
-                        
+
                 if o_dict["llm_factory"] not in res:
                     res[o_dict["llm_factory"]] = {
                         "tags": factory_tags,
                         "llm": []
                     }
-                
+
                 res[o_dict["llm_factory"]]["llm"].append({
                     "type": o_dict["model_type"],
                     "name": o_dict["llm_name"],
@@ -353,14 +356,12 @@ def my_llms():
                     "name": o["llm_name"],
                     "used_token": o["used_tokens"]
                 })
-        
+
         return get_json_result(data=res)
     except Exception as e:
         return server_error_response(e)
 
 
-
-
 @manager.route('/list', methods=['GET'])  # noqa: F821
 @login_required
 def list_app():

api/apps/sdk/chat.py

@@ -21,7 +21,7 @@ from api import settings
 from api.db import StatusEnum
 from api.db.services.dialog_service import DialogService
 from api.db.services.knowledgebase_service import KnowledgebaseService
-from api.db.services.llm_service import TenantLLMService
+from api.db.services.tenant_llm_service import TenantLLMService
 from api.db.services.user_service import TenantService
 from api.utils import get_uuid
 from api.utils.api_utils import check_duplicate_ids, get_error_data_result, get_result, token_required
@@ -99,7 +99,7 @@ def create(tenant_id):
       Here is the knowledge base:
       {knowledge}
       The above is the knowledge base.""",
-        "prologue": "Hi! I'm your assistant, what can I do for you?",
+        "prologue": "Hi! I'm your assistant. What can I do for you?",
         "parameters": [{"key": "knowledge", "optional": False}],
         "empty_response": "Sorry! No relevant content was found in the knowledge base!",
         "quote": True,
@@ -139,7 +139,7 @@ def create(tenant_id):
     res["llm"] = res.pop("llm_setting")
     res["llm"]["model_name"] = res.pop("llm_id")
     del res["kb_ids"]
-    res["dataset_ids"] = req["dataset_ids"]
+    res["dataset_ids"] = req.get("dataset_ids", [])
     res["avatar"] = res.pop("icon")
     return get_result(data=res)
 
@@ -150,10 +150,10 @@ def update(tenant_id, chat_id):
     if not DialogService.query(tenant_id=tenant_id, id=chat_id, status=StatusEnum.VALID.value):
         return get_error_data_result(message="You do not own the chat")
     req = request.json
-    ids = req.get("dataset_ids")
+    ids = req.get("dataset_ids", [])
     if "show_quotation" in req:
         req["do_refer"] = req.pop("show_quotation")
-    if ids is not None:
+    if ids:
         for kb_id in ids:
             kbs = KnowledgebaseService.accessible(kb_id=kb_id, user_id=tenant_id)
             if not kbs:

api/apps/sdk/dify_retrieval.py

@@ -1,4 +1,4 @@
-#
+    #
 #  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
 #
 #  Licensed under the Apache License, Version 2.0 (the "License");
@@ -13,6 +13,8 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 #
+import logging
+
 from flask import request, jsonify
 
 from api.db import LLMType
@@ -22,6 +24,7 @@ from api.db.services.llm_service import LLMBundle
 from api import settings
 from api.utils.api_utils import validate_request, build_error_result, apikey_required
 from rag.app.tag import label_question
+from api.db.services.dialog_service import meta_filter
 
 
 @manager.route('/dify/retrieval', methods=['POST'])  # noqa: F821
@@ -35,18 +38,23 @@ def retrieval(tenant_id):
     retrieval_setting = req.get("retrieval_setting", {})
     similarity_threshold = float(retrieval_setting.get("score_threshold", 0.0))
     top = int(retrieval_setting.get("top_k", 1024))
-
+    metadata_condition = req.get("metadata_condition",{})
+    metas = DocumentService.get_meta_by_kbs([kb_id])
+ 
+    doc_ids = []
     try:
 
         e, kb = KnowledgebaseService.get_by_id(kb_id)
         if not e:
             return build_error_result(message="Knowledgebase not found!", code=settings.RetCode.NOT_FOUND)
 
-        if kb.tenant_id != tenant_id:
-            return build_error_result(message="Knowledgebase not found!", code=settings.RetCode.NOT_FOUND)
-
         embd_mdl = LLMBundle(kb.tenant_id, LLMType.EMBEDDING.value, llm_name=kb.embd_id)
-
+        print(metadata_condition)
+        print("after",convert_conditions(metadata_condition))
+        doc_ids.extend(meta_filter(metas, convert_conditions(metadata_condition)))
+        print("doc_ids",doc_ids)
+        if not doc_ids and metadata_condition is not None:
+            doc_ids = ['-999']
         ranks = settings.retrievaler.retrieval(
             question,
             embd_mdl,
@@ -57,6 +65,7 @@ def retrieval(tenant_id):
             similarity_threshold=similarity_threshold,
             vector_similarity_weight=0.3,
             top=top,
+            doc_ids=doc_ids,
             rank_feature=label_question(question, [kb])
         )
 
@@ -65,6 +74,7 @@ def retrieval(tenant_id):
                                                    [tenant_id],
                                                    [kb_id],
                                                    embd_mdl,
+                                                   doc_ids,
                                                    LLMBundle(kb.tenant_id, LLMType.CHAT))
             if ck["content_with_weight"]:
                 ranks["chunks"].insert(0, ck)
@@ -73,11 +83,13 @@ def retrieval(tenant_id):
         for c in ranks["chunks"]:
             e, doc = DocumentService.get_by_id( c["doc_id"])
             c.pop("vector", None)
+            meta = getattr(doc, 'meta_fields', {})
+            meta["doc_id"] = c["doc_id"]
             records.append({
                 "content": c["content_with_weight"],
                 "score": c["similarity"],
                 "title": c["docnm_kwd"],
-                "metadata": doc.meta_fields
+                "metadata": meta
             })
 
         return jsonify({"records": records})
@@ -87,4 +99,22 @@ def retrieval(tenant_id):
                 message='No chunk found! Check the chunk status please!',
                 code=settings.RetCode.NOT_FOUND
             )
+        logging.exception(e)
         return build_error_result(message=str(e), code=settings.RetCode.SERVER_ERROR)
+
+def convert_conditions(metadata_condition):
+    if metadata_condition is None:
+        metadata_condition = {}
+    op_mapping = {
+        "is": "=",
+        "not is": "≠"
+    }
+    return [
+    {
+        "op": op_mapping.get(cond["comparison_operator"], cond["comparison_operator"]),
+        "key": cond["name"],
+        "value": cond["value"]
+    }
+    for cond in metadata_condition.get("conditions", [])
+]
+

api/apps/sdk/doc.py

@@ -32,13 +32,14 @@ from api.db.services.document_service import DocumentService
 from api.db.services.file2document_service import File2DocumentService
 from api.db.services.file_service import FileService
 from api.db.services.knowledgebase_service import KnowledgebaseService
-from api.db.services.llm_service import LLMBundle, TenantLLMService
+from api.db.services.llm_service import LLMBundle
+from api.db.services.tenant_llm_service import TenantLLMService
 from api.db.services.task_service import TaskService, queue_tasks
 from api.utils.api_utils import check_duplicate_ids, construct_json_result, get_error_data_result, get_parser_config, get_result, server_error_response, token_required
 from rag.app.qa import beAdoc, rmPrefix
 from rag.app.tag import label_question
 from rag.nlp import rag_tokenizer, search
-from rag.prompts import keyword_extraction, cross_languages
+from rag.prompts import cross_languages, keyword_extraction
 from rag.utils import rmSpace
 from rag.utils.storage_factory import STORAGE_IMPL
 
@@ -456,6 +457,18 @@ def list_docs(dataset_id, tenant_id):
         required: false
         default: true
         description: Order in descending.
+    - in: query
+        name: create_time_from
+        type: integer
+        required: false
+        default: 0
+        description: Unix timestamp for filtering documents created after this time. 0 means no filter.
+      - in: query
+        name: create_time_to
+        type: integer
+        required: false
+        default: 0
+        description: Unix timestamp for filtering documents created before this time. 0 means no filter.
       - in: header
         name: Authorization
         type: string
@@ -517,6 +530,17 @@ def list_docs(dataset_id, tenant_id):
         desc = True
     docs, tol = DocumentService.get_list(dataset_id, page, page_size, orderby, desc, keywords, id, name)
 
+    create_time_from = int(request.args.get("create_time_from", 0))
+    create_time_to = int(request.args.get("create_time_to", 0))
+
+    if create_time_from or create_time_to:
+        filtered_docs = []
+        for doc in docs:
+            doc_create_time = doc.get("create_time", 0)
+            if (create_time_from == 0 or doc_create_time >= create_time_from) and (create_time_to == 0 or doc_create_time <= create_time_to):
+                filtered_docs.append(doc)
+        docs = filtered_docs
+
     # rename key's name
     renamed_doc_list = []
     key_mapping = {

api/apps/sdk/session.py

@@ -21,6 +21,7 @@ import tiktoken
 from flask import Response, jsonify, request
 
 from agent.canvas import Canvas
+from api import settings
 from api.db import LLMType, StatusEnum
 from api.db.db_models import APIToken
 from api.db.services.api_service import API4ConversationService
@@ -28,13 +29,18 @@ from api.db.services.canvas_service import UserCanvasService, completionOpenAI
 from api.db.services.canvas_service import completion as agent_completion
 from api.db.services.conversation_service import ConversationService, iframe_completion
 from api.db.services.conversation_service import completion as rag_completion
-from api.db.services.dialog_service import DialogService, ask, chat
-from api.db.services.file_service import FileService
+from api.db.services.dialog_service import DialogService, ask, chat, gen_mindmap, meta_filter
+from api.db.services.document_service import DocumentService
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.llm_service import LLMBundle
+from api.db.services.search_service import SearchService
+from api.db.services.user_service import UserTenantService
 from api.utils import get_uuid
-from api.utils.api_utils import check_duplicate_ids, get_data_openai, get_error_data_result, get_result, token_required, validate_request
+from api.utils.api_utils import check_duplicate_ids, get_data_openai, get_error_data_result, get_json_result, get_result, server_error_response, token_required, validate_request
+from rag.app.tag import label_question
 from rag.prompts import chunks_format
+from rag.prompts.prompt_template import load_prompt
+from rag.prompts.prompts import cross_languages, gen_meta_filter, keyword_extraction
 
 
 @manager.route("/chats/<chat_id>/sessions", methods=["POST"])  # noqa: F821
@@ -51,6 +57,7 @@ def create(tenant_id, chat_id):
         "name": req.get("name", "New session"),
         "message": [{"role": "assistant", "content": dia[0].prompt_config.get("prologue")}],
         "user_id": req.get("user_id", ""),
+        "reference": [{}],
     }
     if not conv.get("name"):
         return get_error_data_result(message="`name` can not be empty.")
@@ -68,11 +75,7 @@ def create(tenant_id, chat_id):
 @manager.route("/agents/<agent_id>/sessions", methods=["POST"])  # noqa: F821
 @token_required
 def create_agent_session(tenant_id, agent_id):
-    req = request.json
-    if not request.is_json:
-        req = request.form
-    files = request.files
-    user_id = request.args.get("user_id", "")
+    user_id = request.args.get("user_id", tenant_id)
     e, cvs = UserCanvasService.get_by_id(agent_id)
     if not e:
         return get_error_data_result("Agent not found.")
@@ -81,45 +84,12 @@ def create_agent_session(tenant_id, agent_id):
     if not isinstance(cvs.dsl, str):
         cvs.dsl = json.dumps(cvs.dsl, ensure_ascii=False)
 
-    canvas = Canvas(cvs.dsl, tenant_id)
+    session_id = get_uuid()
+    canvas = Canvas(cvs.dsl, tenant_id, agent_id)
     canvas.reset()
-    query = canvas.get_preset_param()
-    if query:
-        for ele in query:
-            if not ele["optional"]:
-                if ele["type"] == "file":
-                    if files is None or not files.get(ele["key"]):
-                        return get_error_data_result(f"`{ele['key']}` with type `{ele['type']}` is required")
-                    upload_file = files.get(ele["key"])
-                    file_content = FileService.parse_docs([upload_file], user_id)
-                    file_name = upload_file.filename
-                    ele["value"] = file_name + "\n" + file_content
-                else:
-                    if req is None or not req.get(ele["key"]):
-                        return get_error_data_result(f"`{ele['key']}` with type `{ele['type']}` is required")
-                    ele["value"] = req[ele["key"]]
-            else:
-                if ele["type"] == "file":
-                    if files is not None and files.get(ele["key"]):
-                        upload_file = files.get(ele["key"])
-                        file_content = FileService.parse_docs([upload_file], user_id)
-                        file_name = upload_file.filename
-                        ele["value"] = file_name + "\n" + file_content
-                    else:
-                        if "value" in ele:
-                            ele.pop("value")
-                else:
-                    if req is not None and req.get(ele["key"]):
-                        ele["value"] = req[ele["key"]]
-                    else:
-                        if "value" in ele:
-                            ele.pop("value")
-
-    for ans in canvas.run(stream=False):
-        pass
 
     cvs.dsl = json.loads(str(canvas))
-    conv = {"id": get_uuid(), "dialog_id": cvs.id, "user_id": user_id, "message": [{"role": "assistant", "content": canvas.get_prologue()}], "source": "agent", "dsl": cvs.dsl}
+    conv = {"id": session_id, "dialog_id": cvs.id, "user_id": user_id, "message": [{"role": "assistant", "content": canvas.get_prologue()}], "source": "agent", "dsl": cvs.dsl}
     API4ConversationService.save(**conv)
     conv["agent_id"] = conv.pop("dialog_id")
     return get_result(data=conv)
@@ -435,14 +405,38 @@ def agents_completion_openai_compatibility(tenant_id, agent_id):
             )
         )
 
-    # Get the last user message as the question
     question = next((m["content"] for m in reversed(messages) if m["role"] == "user"), "")
 
-    if req.get("stream", True):
-        return Response(completionOpenAI(tenant_id, agent_id, question, session_id=req.get("id", req.get("metadata", {}).get("id", "")), stream=True), mimetype="text/event-stream")
+    stream = req.pop("stream", False)
+    if stream:
+        resp = Response(
+            completionOpenAI(
+                tenant_id,
+                agent_id,
+                question,
+                session_id=req.get("id", req.get("metadata", {}).get("id", "")),
+                stream=True,
+                **req,
+            ),
+            mimetype="text/event-stream",
+        )
+        resp.headers.add_header("Cache-control", "no-cache")
+        resp.headers.add_header("Connection", "keep-alive")
+        resp.headers.add_header("X-Accel-Buffering", "no")
+        resp.headers.add_header("Content-Type", "text/event-stream; charset=utf-8")
+        return resp
     else:
         # For non-streaming, just return the response directly
-        response = next(completionOpenAI(tenant_id, agent_id, question, session_id=req.get("id", req.get("metadata", {}).get("id", "")), stream=False))
+        response = next(
+            completionOpenAI(
+                tenant_id,
+                agent_id,
+                question,
+                session_id=req.get("id", req.get("metadata", {}).get("id", "")),
+                stream=False,
+                **req,
+            )
+        )
         return jsonify(response)
 
 
@@ -450,41 +444,47 @@ def agents_completion_openai_compatibility(tenant_id, agent_id):
 @token_required
 def agent_completions(tenant_id, agent_id):
     req = request.json
-    cvs = UserCanvasService.query(user_id=tenant_id, id=agent_id)
-    if not cvs:
-        return get_error_data_result(f"You don't own the agent {agent_id}")
-    if req.get("session_id"):
-        dsl = cvs[0].dsl
-        if not isinstance(dsl, str):
-            dsl = json.dumps(dsl)
 
-        conv = API4ConversationService.query(id=req["session_id"], dialog_id=agent_id)
-        if not conv:
-            return get_error_data_result(f"You don't own the session {req['session_id']}")
-        # If an update to UserCanvas is detected, update the API4Conversation.dsl
-        sync_dsl = req.get("sync_dsl", False)
-        if sync_dsl is True and cvs[0].update_time > conv[0].update_time:
-            current_dsl = conv[0].dsl
-            new_dsl = json.loads(dsl)
-            state_fields = ["history", "messages", "path", "reference"]
-            states = {field: current_dsl.get(field, []) for field in state_fields}
-            current_dsl.update(new_dsl)
-            current_dsl.update(states)
-            API4ConversationService.update_by_id(req["session_id"], {"dsl": current_dsl})
-    else:
-        req["question"] = ""
+    ans = {}
     if req.get("stream", True):
-        resp = Response(agent_completion(tenant_id, agent_id, **req), mimetype="text/event-stream")
+
+        def generate():
+            for answer in agent_completion(tenant_id=tenant_id, agent_id=agent_id, **req):
+                if isinstance(answer, str):
+                    try:
+                        ans = json.loads(answer[5:])  # remove "data:"
+                    except Exception:
+                        continue
+
+                if ans.get("event") != "message" or not ans.get("data", {}).get("reference", None):
+                    continue
+
+                yield answer
+
+            yield "data:[DONE]\n\n"
+
+    if req.get("stream", True):
+        resp = Response(generate(), mimetype="text/event-stream")
         resp.headers.add_header("Cache-control", "no-cache")
         resp.headers.add_header("Connection", "keep-alive")
         resp.headers.add_header("X-Accel-Buffering", "no")
         resp.headers.add_header("Content-Type", "text/event-stream; charset=utf-8")
         return resp
-    try:
-        for answer in agent_completion(tenant_id, agent_id, **req):
-            return get_result(data=answer)
-    except Exception as e:
-        return get_error_data_result(str(e))
+
+    full_content = ""
+    for answer in agent_completion(tenant_id=tenant_id, agent_id=agent_id, **req):
+        try:
+            ans = json.loads(answer[5:])
+
+            if ans["event"] == "message":
+                full_content += ans["data"]["content"]
+
+            if ans.get("data", {}).get("reference", None):
+                ans["data"]["content"] = full_content
+                return get_result(data=ans)
+        except Exception as e:
+            return get_result(data=f"**ERROR**: {str(e)}")
+    return get_result(data=ans)
 
 
 @manager.route("/chats/<chat_id>/sessions", methods=["GET"])  # noqa: F821
@@ -512,16 +512,16 @@ def list_session(tenant_id, chat_id):
             if "prompt" in info:
                 info.pop("prompt")
         conv["chat_id"] = conv.pop("dialog_id")
-        if conv["reference"]:
+        ref_messages = conv["reference"]
+        if ref_messages:
             messages = conv["messages"]
             message_num = 0
-            while message_num < len(messages) and message_num < len(conv["reference"]):
-                if message_num != 0 and messages[message_num]["role"] != "user":
-                    if message_num >= len(conv["reference"]):
-                        break
+            ref_num = 0
+            while message_num < len(messages) and ref_num < len(ref_messages):
+                if messages[message_num]["role"] != "user":
                     chunk_list = []
-                    if "chunks" in conv["reference"][message_num]:
-                        chunks = conv["reference"][message_num]["chunks"]
+                    if "chunks" in ref_messages[ref_num]:
+                        chunks = ref_messages[ref_num]["chunks"]
                         for chunk in chunks:
                             new_chunk = {
                                 "id": chunk.get("chunk_id", chunk.get("id")),
@@ -535,6 +535,7 @@ def list_session(tenant_id, chat_id):
 
                             chunk_list.append(new_chunk)
                     messages[message_num]["reference"] = chunk_list
+                    ref_num += 1
                 message_num += 1
         del conv["reference"]
     return get_result(data=convs)
@@ -566,16 +567,24 @@ def list_agent_session(tenant_id, agent_id):
             if "prompt" in info:
                 info.pop("prompt")
         conv["agent_id"] = conv.pop("dialog_id")
+        # Fix for session listing endpoint
         if conv["reference"]:
             messages = conv["messages"]
             message_num = 0
             chunk_num = 0
+            # Ensure reference is a list type to prevent KeyError
+            if not isinstance(conv["reference"], list):
+                conv["reference"] = []
             while message_num < len(messages):
                 if message_num != 0 and messages[message_num]["role"] != "user":
                     chunk_list = []
-                    if "chunks" in conv["reference"][chunk_num]:
+                    # Add boundary and type checks to prevent KeyError
+                    if chunk_num < len(conv["reference"]) and conv["reference"][chunk_num] is not None and isinstance(conv["reference"][chunk_num], dict) and "chunks" in conv["reference"][chunk_num]:
                         chunks = conv["reference"][chunk_num]["chunks"]
                         for chunk in chunks:
+                            # Ensure chunk is a dictionary before calling get method
+                            if not isinstance(chunk, dict):
+                                continue
                             new_chunk = {
                                 "id": chunk.get("chunk_id", chunk.get("id")),
                                 "content": chunk.get("content_with_weight", chunk.get("content")),
@@ -809,6 +818,29 @@ def chatbot_completions(dialog_id):
         return get_result(data=answer)
 
 
+@manager.route("/chatbots/<dialog_id>/info", methods=["GET"])  # noqa: F821
+def chatbots_inputs(dialog_id):
+    token = request.headers.get("Authorization").split()
+    if len(token) != 2:
+        return get_error_data_result(message='Authorization is not valid!"')
+    token = token[1]
+    objs = APIToken.query(beta=token)
+    if not objs:
+        return get_error_data_result(message='Authentication error: API key is invalid!"')
+
+    e, dialog = DialogService.get_by_id(dialog_id)
+    if not e:
+        return get_error_data_result(f"Can't find dialog by ID: {dialog_id}")
+
+    return get_result(
+        data={
+            "title": dialog.name,
+            "avatar": dialog.icon,
+            "prologue": dialog.prompt_config.get("prologue", ""),
+        }
+    )
+
+
 @manager.route("/agentbots/<agent_id>/completions", methods=["POST"])  # noqa: F821
 def agent_bot_completions(agent_id):
     req = request.json
@@ -848,10 +880,231 @@ def begin_inputs(agent_id):
         return get_error_data_result(f"Can't find agent by ID: {agent_id}")
 
     canvas = Canvas(json.dumps(cvs.dsl), objs[0].tenant_id)
-    return get_result(data={
-        "title": cvs.title,
-        "avatar": cvs.avatar,
-        "inputs": canvas.get_component_input_form("begin")
-    })
+    return get_result(data={"title": cvs.title, "avatar": cvs.avatar, "inputs": canvas.get_component_input_form("begin"), "prologue": canvas.get_prologue(), "mode": canvas.get_mode()})
 
 
+@manager.route("/searchbots/ask", methods=["POST"])  # noqa: F821
+@validate_request("question", "kb_ids")
+def ask_about_embedded():
+    token = request.headers.get("Authorization").split()
+    if len(token) != 2:
+        return get_error_data_result(message='Authorization is not valid!"')
+    token = token[1]
+    objs = APIToken.query(beta=token)
+    if not objs:
+        return get_error_data_result(message='Authentication error: API key is invalid!"')
+
+    req = request.json
+    uid = objs[0].tenant_id
+
+    search_id = req.get("search_id", "")
+    search_config = {}
+    if search_id:
+        if search_app := SearchService.get_detail(search_id):
+            search_config = search_app.get("search_config", {})
+
+    def stream():
+        nonlocal req, uid
+        try:
+            for ans in ask(req["question"], req["kb_ids"], uid, search_config=search_config):
+                yield "data:" + json.dumps({"code": 0, "message": "", "data": ans}, ensure_ascii=False) + "\n\n"
+        except Exception as e:
+            yield "data:" + json.dumps({"code": 500, "message": str(e), "data": {"answer": "**ERROR**: " + str(e), "reference": []}}, ensure_ascii=False) + "\n\n"
+        yield "data:" + json.dumps({"code": 0, "message": "", "data": True}, ensure_ascii=False) + "\n\n"
+
+    resp = Response(stream(), mimetype="text/event-stream")
+    resp.headers.add_header("Cache-control", "no-cache")
+    resp.headers.add_header("Connection", "keep-alive")
+    resp.headers.add_header("X-Accel-Buffering", "no")
+    resp.headers.add_header("Content-Type", "text/event-stream; charset=utf-8")
+    return resp
+
+
+@manager.route("/searchbots/retrieval_test", methods=["POST"])  # noqa: F821
+@validate_request("kb_id", "question")
+def retrieval_test_embedded():
+    token = request.headers.get("Authorization").split()
+    if len(token) != 2:
+        return get_error_data_result(message='Authorization is not valid!"')
+    token = token[1]
+    objs = APIToken.query(beta=token)
+    if not objs:
+        return get_error_data_result(message='Authentication error: API key is invalid!"')
+
+    req = request.json
+    page = int(req.get("page", 1))
+    size = int(req.get("size", 30))
+    question = req["question"]
+    kb_ids = req["kb_id"]
+    if isinstance(kb_ids, str):
+        kb_ids = [kb_ids]
+    doc_ids = req.get("doc_ids", [])
+    similarity_threshold = float(req.get("similarity_threshold", 0.0))
+    vector_similarity_weight = float(req.get("vector_similarity_weight", 0.3))
+    use_kg = req.get("use_kg", False)
+    top = int(req.get("top_k", 1024))
+    langs = req.get("cross_languages", [])
+    tenant_ids = []
+
+    tenant_id = objs[0].tenant_id
+    if not tenant_id:
+        return get_error_data_result(message="permission denined.")
+
+    if req.get("search_id", ""):
+        search_config = SearchService.get_detail(req.get("search_id", "")).get("search_config", {})
+        meta_data_filter = search_config.get("meta_data_filter", {})
+        metas = DocumentService.get_meta_by_kbs(kb_ids)
+        if meta_data_filter.get("method") == "auto":
+            chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, llm_name=search_config.get("chat_id", ""))
+            filters = gen_meta_filter(chat_mdl, metas, question)
+            doc_ids.extend(meta_filter(metas, filters))
+            if not doc_ids:
+                doc_ids = None
+        elif meta_data_filter.get("method") == "manual":
+            doc_ids.extend(meta_filter(metas, meta_data_filter["manual"]))
+            if not doc_ids:
+                doc_ids = None
+
+    try:
+        tenants = UserTenantService.query(user_id=tenant_id)
+        for kb_id in kb_ids:
+            for tenant in tenants:
+                if KnowledgebaseService.query(tenant_id=tenant.tenant_id, id=kb_id):
+                    tenant_ids.append(tenant.tenant_id)
+                    break
+            else:
+                return get_json_result(data=False, message="Only owner of knowledgebase authorized for this operation.", code=settings.RetCode.OPERATING_ERROR)
+
+        e, kb = KnowledgebaseService.get_by_id(kb_ids[0])
+        if not e:
+            return get_error_data_result(message="Knowledgebase not found!")
+
+        if langs:
+            question = cross_languages(kb.tenant_id, None, question, langs)
+
+        embd_mdl = LLMBundle(kb.tenant_id, LLMType.EMBEDDING.value, llm_name=kb.embd_id)
+
+        rerank_mdl = None
+        if req.get("rerank_id"):
+            rerank_mdl = LLMBundle(kb.tenant_id, LLMType.RERANK.value, llm_name=req["rerank_id"])
+
+        if req.get("keyword", False):
+            chat_mdl = LLMBundle(kb.tenant_id, LLMType.CHAT)
+            question += keyword_extraction(chat_mdl, question)
+
+        labels = label_question(question, [kb])
+        ranks = settings.retrievaler.retrieval(
+            question, embd_mdl, tenant_ids, kb_ids, page, size, similarity_threshold, vector_similarity_weight, top, doc_ids, rerank_mdl=rerank_mdl, highlight=req.get("highlight"), rank_feature=labels
+        )
+        if use_kg:
+            ck = settings.kg_retrievaler.retrieval(question, tenant_ids, kb_ids, embd_mdl, LLMBundle(kb.tenant_id, LLMType.CHAT))
+            if ck["content_with_weight"]:
+                ranks["chunks"].insert(0, ck)
+
+        for c in ranks["chunks"]:
+            c.pop("vector", None)
+        ranks["labels"] = labels
+
+        return get_json_result(data=ranks)
+    except Exception as e:
+        if str(e).find("not_found") > 0:
+            return get_json_result(data=False, message="No chunk found! Check the chunk status please!", code=settings.RetCode.DATA_ERROR)
+        return server_error_response(e)
+
+
+@manager.route("/searchbots/related_questions", methods=["POST"])  # noqa: F821
+@validate_request("question")
+def related_questions_embedded():
+    token = request.headers.get("Authorization").split()
+    if len(token) != 2:
+        return get_error_data_result(message='Authorization is not valid!"')
+    token = token[1]
+    objs = APIToken.query(beta=token)
+    if not objs:
+        return get_error_data_result(message='Authentication error: API key is invalid!"')
+
+    req = request.json
+    tenant_id = objs[0].tenant_id
+    if not tenant_id:
+        return get_error_data_result(message="permission denined.")
+
+    search_id = req.get("search_id", "")
+    search_config = {}
+    if search_id:
+        if search_app := SearchService.get_detail(search_id):
+            search_config = search_app.get("search_config", {})
+
+    question = req["question"]
+
+    chat_id = search_config.get("chat_id", "")
+    chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, chat_id)
+
+    gen_conf = search_config.get("llm_setting", {"temperature": 0.9})
+    prompt = load_prompt("related_question")
+    ans = chat_mdl.chat(
+        prompt,
+        [
+            {
+                "role": "user",
+                "content": f"""
+Keywords: {question}
+Related search terms:
+    """,
+            }
+        ],
+        gen_conf,
+    )
+    return get_json_result(data=[re.sub(r"^[0-9]\. ", "", a) for a in ans.split("\n") if re.match(r"^[0-9]\. ", a)])
+
+
+@manager.route("/searchbots/detail", methods=["GET"])  # noqa: F821
+def detail_share_embedded():
+    token = request.headers.get("Authorization").split()
+    if len(token) != 2:
+        return get_error_data_result(message='Authorization is not valid!"')
+    token = token[1]
+    objs = APIToken.query(beta=token)
+    if not objs:
+        return get_error_data_result(message='Authentication error: API key is invalid!"')
+
+    search_id = request.args["search_id"]
+    tenant_id = objs[0].tenant_id
+    if not tenant_id:
+        return get_error_data_result(message="permission denined.")
+    try:
+        tenants = UserTenantService.query(user_id=tenant_id)
+        for tenant in tenants:
+            if SearchService.query(tenant_id=tenant.tenant_id, id=search_id):
+                break
+        else:
+            return get_json_result(data=False, message="Has no permission for this operation.", code=settings.RetCode.OPERATING_ERROR)
+
+        search = SearchService.get_detail(search_id)
+        if not search:
+            return get_error_data_result(message="Can't find this Search App!")
+        return get_json_result(data=search)
+    except Exception as e:
+        return server_error_response(e)
+
+
+@manager.route("/searchbots/mindmap", methods=["POST"])  # noqa: F821
+@validate_request("question", "kb_ids")
+def mindmap():
+    token = request.headers.get("Authorization").split()
+    if len(token) != 2:
+        return get_error_data_result(message='Authorization is not valid!"')
+    token = token[1]
+    objs = APIToken.query(beta=token)
+    if not objs:
+        return get_error_data_result(message='Authentication error: API key is invalid!"')
+
+    tenant_id = objs[0].tenant_id
+    req = request.json
+
+    search_id = req.get("search_id", "")
+    search_app = SearchService.get_detail(search_id) if search_id else {}
+
+    mind_map = gen_mindmap(req["question"], req["kb_ids"], tenant_id, search_app.get("search_config", {}))
+    if "error" in mind_map:
+        return server_error_response(Exception(mind_map["error"]))
+    return get_json_result(data=mind_map)

api/apps/search_app.py

@@ -22,7 +22,6 @@ from api.constants import DATASET_NAME_LIMIT
 from api.db import StatusEnum
 from api.db.db_models import DB
 from api.db.services import duplicate_name
-from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.search_service import SearchService
 from api.db.services.user_service import TenantService, UserTenantService
 from api.utils import get_uuid
@@ -47,7 +46,7 @@ def create():
         return get_data_error_result(message="Authorizationd identity.")
 
     search_name = search_name.strip()
-    search_name = duplicate_name(KnowledgebaseService.query, name=search_name, tenant_id=current_user.id, status=StatusEnum.VALID.value)
+    search_name = duplicate_name(SearchService.query, name=search_name, tenant_id=current_user.id, status=StatusEnum.VALID.value)
 
     req["id"] = get_uuid()
     req["name"] = search_name
@@ -156,8 +155,9 @@ def list_search_app():
     owner_ids = req.get("owner_ids", [])
     try:
         if not owner_ids:
-            tenants = TenantService.get_joined_tenants_by_user_id(current_user.id)
-            tenants = [m["tenant_id"] for m in tenants]
+            # tenants = TenantService.get_joined_tenants_by_user_id(current_user.id)
+            # tenants = [m["tenant_id"] for m in tenants]
+            tenants = []
             search_apps, total = SearchService.get_by_tenant_ids(tenants, current_user.id, page_number, items_per_page, orderby, desc, keywords)
         else:
             tenants = owner_ids

api/apps/tenant_app.py

@@ -18,12 +18,14 @@ from flask import request
 from flask_login import login_required, current_user
 
 from api import settings
+from api.apps import smtp_mail_server
 from api.db import UserTenantRole, StatusEnum
 from api.db.db_models import UserTenant
 from api.db.services.user_service import UserTenantService, UserService
 
 from api.utils import get_uuid, delta_seconds
 from api.utils.api_utils import get_json_result, validate_request, server_error_response, get_data_error_result
+from api.utils.web_utils import send_invite_email
 
 
 @manager.route("/<tenant_id>/user/list", methods=["GET"])  # noqa: F821
@@ -78,6 +80,20 @@ def create(tenant_id):
         role=UserTenantRole.INVITE,
         status=StatusEnum.VALID.value)
 
+    if smtp_mail_server and settings.SMTP_CONF:
+        from threading import Thread
+
+        user_name = ""
+        _, user = UserService.get_by_id(current_user.id)
+        if user:
+            user_name = user.nickname
+
+        Thread(
+            target=send_invite_email,
+            args=(invite_user_email, settings.MAIL_FRONTEND_URL, tenant_id, user_name or current_user.email),
+            daemon=True
+        ).start()
+
     usr = invite_users[0].to_dict()
     usr = {k: v for k, v in usr.items() if k in ["id", "avatar", "email", "nickname"]}
 

api/apps/user_app.py

@@ -28,7 +28,8 @@ from api.apps.auth import get_auth_client
 from api.db import FileType, UserTenantRole
 from api.db.db_models import TenantLLM
 from api.db.services.file_service import FileService
-from api.db.services.llm_service import LLMService, TenantLLMService
+from api.db.services.llm_service import get_init_tenant_llm
+from api.db.services.tenant_llm_service import TenantLLMService
 from api.db.services.user_service import TenantService, UserService, UserTenantService
 from api.utils import (
     current_timestamp,
@@ -619,33 +620,8 @@ def user_register(user_id, user):
         "size": 0,
         "location": "",
     }
-    tenant_llm = []
-    for llm in LLMService.query(fid=settings.LLM_FACTORY):
-        tenant_llm.append(
-            {
-                "tenant_id": user_id,
-                "llm_factory": settings.LLM_FACTORY,
-                "llm_name": llm.llm_name,
-                "model_type": llm.model_type,
-                "api_key": settings.API_KEY,
-                "api_base": settings.LLM_BASE_URL,
-                "max_tokens": llm.max_tokens if llm.max_tokens else 8192,
-            }
-        )
-    if settings.LIGHTEN != 1:
-        for buildin_embedding_model in settings.BUILTIN_EMBEDDING_MODELS:
-            mdlnm, fid = TenantLLMService.split_model_name_and_factory(buildin_embedding_model)
-            tenant_llm.append(
-                {
-                    "tenant_id": user_id,
-                    "llm_factory": fid,
-                    "llm_name": mdlnm,
-                    "model_type": "embedding",
-                    "api_key": "",
-                    "api_base": "",
-                    "max_tokens": 1024 if buildin_embedding_model == "BAAI/bge-large-zh-v1.5@BAAI" else 512,
-                }
-            )
+
+    tenant_llm = get_init_tenant_llm(user_id)
 
     if not UserService.save(**user):
         return

api/db/db_models.py

@@ -742,8 +742,9 @@ class Dialog(DataBaseModel):
     prompt_type = CharField(max_length=16, null=False, default="simple", help_text="simple|advanced", index=True)
     prompt_config = JSONField(
         null=False,
-        default={"system": "", "prologue": "Hi! I'm your assistant, what can I do for you?", "parameters": [], "empty_response": "Sorry! No relevant content was found in the knowledge base!"},
+        default={"system": "", "prologue": "Hi! I'm your assistant. What can I do for you?", "parameters": [], "empty_response": "Sorry! No relevant content was found in the knowledge base!"},
     )
+    meta_data_filter = JSONField(null=True, default={})
 
     similarity_threshold = FloatField(default=0.2)
     vector_similarity_weight = FloatField(default=0.3)
@@ -871,7 +872,7 @@ class Search(DataBaseModel):
         default={
             "kb_ids": [],
             "doc_ids": [],
-            "similarity_threshold": 0.0,
+            "similarity_threshold": 0.2,
             "vector_similarity_weight": 0.3,
             "use_kg": False,
             # rerank settings
@@ -880,11 +881,12 @@ class Search(DataBaseModel):
             # chat settings
             "summary": False,
             "chat_id": "",
+            # Leave it here for reference, don't need to set default values
             "llm_setting": {
-                "temperature": 0.1,
-                "top_p": 0.3,
-                "frequency_penalty": 0.7,
-                "presence_penalty": 0.4,
+                # "temperature": 0.1,
+                # "top_p": 0.3,
+                # "frequency_penalty": 0.7,
+                # "presence_penalty": 0.4,
             },
             "chat_settingcross_languages": [],
             "highlight": False,
@@ -1015,4 +1017,8 @@ def migrate_db():
         migrate(migrator.add_column("api_4_conversation", "errors", TextField(null=True, help_text="errors")))
     except Exception:
         pass
-    logging.disable(logging.NOTSET)
+    try:
+        migrate(migrator.add_column("dialog", "meta_data_filter", JSONField(null=True, default={})))
+    except Exception:
+        pass
+    logging.disable(logging.NOTSET)

api/db/init_data.py

@@ -27,7 +27,8 @@ from api.db.services import UserService
 from api.db.services.canvas_service import CanvasTemplateService
 from api.db.services.document_service import DocumentService
 from api.db.services.knowledgebase_service import KnowledgebaseService
-from api.db.services.llm_service import LLMFactoriesService, LLMService, TenantLLMService, LLMBundle
+from api.db.services.tenant_llm_service import LLMFactoriesService, TenantLLMService
+from api.db.services.llm_service import LLMService, LLMBundle, get_init_tenant_llm
 from api.db.services.user_service import TenantService, UserTenantService
 from api import settings
 from api.utils.file_utils import get_project_base_directory
@@ -63,12 +64,8 @@ def init_superuser():
         "invited_by": user_info["id"],
         "role": UserTenantRole.OWNER
     }
-    tenant_llm = []
-    for llm in LLMService.query(fid=settings.LLM_FACTORY):
-        tenant_llm.append(
-            {"tenant_id": user_info["id"], "llm_factory": settings.LLM_FACTORY, "llm_name": llm.llm_name,
-             "model_type": llm.model_type,
-             "api_key": settings.API_KEY, "api_base": settings.LLM_BASE_URL})
+
+    tenant_llm = get_init_tenant_llm(user_info["id"])
 
     if not UserService.save(**user_info):
         logging.error("can't init admin.")
@@ -103,7 +100,7 @@ def init_llm_factory():
     except Exception:
         pass
 
-    factory_llm_infos = settings.FACTORY_LLM_INFOS    
+    factory_llm_infos = settings.FACTORY_LLM_INFOS
     for factory_llm_info in factory_llm_infos:
         info = deepcopy(factory_llm_info)
         llm_infos = info.pop("llm")

api/db/services/canvas_service.py

@@ -16,7 +16,6 @@
 import json
 import logging
 import time
-import traceback
 from uuid import uuid4
 from agent.canvas import Canvas
 from api.db import TenantPermission
@@ -54,12 +53,12 @@ class UserCanvasService(CommonService):
         agents = agents.paginate(page_number, items_per_page)
 
         return list(agents.dicts())
-   
+
     @classmethod
     @DB.connection_context()
     def get_by_tenant_id(cls, pid):
         try:
-            
+
             fields = [
                 cls.model.id,
                 cls.model.avatar,
@@ -83,7 +82,7 @@ class UserCanvasService(CommonService):
         except Exception as e:
             logging.exception(e)
             return False, None
-     
+
     @classmethod
     @DB.connection_context()
     def get_by_tenant_ids(cls, joined_tenant_ids, user_id,
@@ -103,14 +102,14 @@ class UserCanvasService(CommonService):
         ]
         if keywords:
             agents = cls.model.select(*fields).join(User, on=(cls.model.user_id == User.id)).where(
-                ((cls.model.user_id.in_(joined_tenant_ids) & (cls.model.permission == 
+                ((cls.model.user_id.in_(joined_tenant_ids) & (cls.model.permission ==
                                                                 TenantPermission.TEAM.value)) | (
                     cls.model.user_id == user_id)),
                 (fn.LOWER(cls.model.title).contains(keywords.lower()))
             )
         else:
             agents = cls.model.select(*fields).join(User, on=(cls.model.user_id == User.id)).where(
-                ((cls.model.user_id.in_(joined_tenant_ids) & (cls.model.permission == 
+                ((cls.model.user_id.in_(joined_tenant_ids) & (cls.model.permission ==
                                                                 TenantPermission.TEAM.value)) | (
                     cls.model.user_id == user_id))
             )
@@ -122,9 +121,22 @@ class UserCanvasService(CommonService):
         agents = agents.paginate(page_number, items_per_page)
         return list(agents.dicts()), count
 
+    @classmethod
+    @DB.connection_context()
+    def accessible(cls, canvas_id, tenant_id):
+        from api.db.services.user_service import UserTenantService
+        e, c = UserCanvasService.get_by_tenant_id(canvas_id)
+        if not e:
+            return False
+
+        tids = [t.tenant_id for t in UserTenantService.query(user_id=tenant_id)]
+        if c["user_id"] != canvas_id and c["user_id"]  not in tids:
+            return False
+        return True
+
 
 def completion(tenant_id, agent_id, session_id=None, **kwargs):
-    query = kwargs.get("query", "")
+    query = kwargs.get("query", "") or kwargs.get("question", "")
     files = kwargs.get("files", [])
     inputs = kwargs.get("inputs", {})
     user_id = kwargs.get("user_id", "")
@@ -152,7 +164,8 @@ def completion(tenant_id, agent_id, session_id=None, **kwargs):
             "user_id": user_id,
             "message": [],
             "source": "agent",
-            "dsl": cvs.dsl
+            "dsl": cvs.dsl,
+            "reference": []
         }
         API4ConversationService.save(**conv)
         conv = API4Conversation(**conv)
@@ -173,223 +186,103 @@ def completion(tenant_id, agent_id, session_id=None, **kwargs):
     conv.message.append({"role": "assistant", "content": txt, "created_at": time.time(), "id": message_id})
     conv.reference = canvas.get_reference()
     conv.errors = canvas.error
-    API4ConversationService.append_message(conv.id, conv.to_dict())
+    conv.dsl = str(canvas)
+    conv = conv.to_dict()
+    API4ConversationService.append_message(conv["id"], conv)
 
 
 def completionOpenAI(tenant_id, agent_id, question, session_id=None, stream=True, **kwargs):
-    """Main function for OpenAI-compatible completions, structured similarly to the completion function."""
     tiktokenenc = tiktoken.get_encoding("cl100k_base")
-    e, cvs = UserCanvasService.get_by_id(agent_id)
-    
-    if not e:
-        yield get_data_openai(
-            id=session_id,
-            model=agent_id,
-            content="**ERROR**: Agent not found."
-        )
-        return
-    
-    if cvs.user_id != tenant_id:
-        yield get_data_openai(
-            id=session_id,
-            model=agent_id,
-            content="**ERROR**: You do not own the agent"
-        )
-        return
-    
-    if not isinstance(cvs.dsl, str):
-        cvs.dsl = json.dumps(cvs.dsl, ensure_ascii=False)
-    
-    canvas = Canvas(cvs.dsl, tenant_id)
-    canvas.reset()
-    message_id = str(uuid4())
-    
-    # Handle new session creation
-    if not session_id:
-        query = canvas.get_preset_param()
-        if query:
-            for ele in query:
-                if not ele["optional"]:
-                    if not kwargs.get(ele["key"]):
-                        yield get_data_openai(
-                            id=None,
-                            model=agent_id,
-                            content=f"`{ele['key']}` is required",
-                            completion_tokens=len(tiktokenenc.encode(f"`{ele['key']}` is required")),
-                            prompt_tokens=len(tiktokenenc.encode(question if question else ""))
-                        )
-                        return
-                    ele["value"] = kwargs[ele["key"]]
-                if ele["optional"]:
-                    if kwargs.get(ele["key"]):
-                        ele["value"] = kwargs[ele['key']]
-                    else:
-                        if "value" in ele:
-                            ele.pop("value")
-        
-        cvs.dsl = json.loads(str(canvas))
-        session_id = get_uuid()
-        conv = {
-            "id": session_id,
-            "dialog_id": cvs.id,
-            "user_id": kwargs.get("user_id", "") if isinstance(kwargs, dict) else "",
-            "message": [{"role": "assistant", "content": canvas.get_prologue(), "created_at": time.time()}],
-            "source": "agent",
-            "dsl": cvs.dsl
-        }
-        canvas.messages.append({"role": "user", "content": question, "id": message_id})
-        canvas.add_user_input(question)
-
-        API4ConversationService.save(**conv)
-        conv = API4Conversation(**conv)
-        if not conv.message:
-            conv.message = []
-        conv.message.append({
-            "role": "user",
-            "content": question,
-            "id": message_id
-        })
-        
-        if not conv.reference:
-            conv.reference = []
-        conv.reference.append({"chunks": [], "doc_aggs": []})
-            
-    # Handle existing session
-    else:
-        e, conv = API4ConversationService.get_by_id(session_id)
-        if not e:
-            yield get_data_openai(
-                id=session_id,
-                model=agent_id,
-                content="**ERROR**: Session not found!"
-            )
-            return
-        
-        canvas = Canvas(json.dumps(conv.dsl), tenant_id)
-        canvas.messages.append({"role": "user", "content": question, "id": message_id})
-        canvas.add_user_input(question)
-        
-        if not conv.message:
-            conv.message = []
-        conv.message.append({
-            "role": "user",
-            "content": question,
-            "id": message_id
-        })
-        
-        if not conv.reference:
-            conv.reference = []
-        conv.reference.append({"chunks": [], "doc_aggs": []})
-    
-    # Process request based on stream mode
-    final_ans = {"reference": [], "content": ""}
     prompt_tokens = len(tiktokenenc.encode(str(question)))
-    
+    user_id = kwargs.get("user_id", "")
+
     if stream:
+        completion_tokens = 0
         try:
-            completion_tokens = 0
-            for ans in canvas.run(stream=True, bypass_begin=True):
-                if ans.get("running_status"):
-                    completion_tokens += len(tiktokenenc.encode(ans.get("content", "")))
-                    yield "data: " + json.dumps(
-                        get_data_openai(
-                            id=session_id,
-                            model=agent_id,
-                            content=ans["content"],
-                            object="chat.completion.chunk",
-                            completion_tokens=completion_tokens,
-                            prompt_tokens=prompt_tokens
-                        ),
-                        ensure_ascii=False
-                    ) + "\n\n"
+            for ans in completion(
+                tenant_id=tenant_id,
+                agent_id=agent_id,
+                session_id=session_id,
+                query=question,
+                user_id=user_id,
+                **kwargs
+            ):
+                if isinstance(ans, str):
+                    try:
+                        ans = json.loads(ans[5:])  # remove "data:"
+                    except Exception as e:
+                        logging.exception(f"Agent OpenAI-Compatible completionOpenAI parse answer failed: {e}")
+                        continue
+                if ans.get("event") != "message" or not ans.get("data", {}).get("reference", None):
                     continue
-                
-                for k in ans.keys():
-                    final_ans[k] = ans[k]
-                
-                completion_tokens += len(tiktokenenc.encode(final_ans.get("content", "")))
+                content_piece = ans["data"]["content"]
+                completion_tokens += len(tiktokenenc.encode(content_piece))
+
                 yield "data: " + json.dumps(
                     get_data_openai(
-                        id=session_id,
+                        id=session_id or str(uuid4()),
                         model=agent_id,
-                        content=final_ans["content"],
-                        object="chat.completion.chunk",
-                        finish_reason="stop",
+                        content=content_piece,
+                        prompt_tokens=prompt_tokens,
                         completion_tokens=completion_tokens,
-                        prompt_tokens=prompt_tokens
+                        stream=True
                     ),
                     ensure_ascii=False
                 ) + "\n\n"
-            
-            # Update conversation
-            canvas.messages.append({"role": "assistant", "content": final_ans["content"], "created_at": time.time(), "id": message_id})
-            canvas.history.append(("assistant", final_ans["content"]))
-            if final_ans.get("reference"):
-                canvas.reference.append(final_ans["reference"])
-            conv.dsl = json.loads(str(canvas))
-            API4ConversationService.append_message(conv.id, conv.to_dict())
-            
+
             yield "data: [DONE]\n\n"
-            
+
         except Exception as e:
-            traceback.print_exc()
-            conv.dsl = json.loads(str(canvas))
-            API4ConversationService.append_message(conv.id, conv.to_dict())
             yield "data: " + json.dumps(
                 get_data_openai(
-                    id=session_id,
+                    id=session_id or str(uuid4()),
                     model=agent_id,
-                    content="**ERROR**: " + str(e),
+                    content=f"**ERROR**: {str(e)}",
                     finish_reason="stop",
-                    completion_tokens=len(tiktokenenc.encode("**ERROR**: " + str(e))),
-                    prompt_tokens=prompt_tokens
+                    prompt_tokens=prompt_tokens,
+                    completion_tokens=len(tiktokenenc.encode(f"**ERROR**: {str(e)}")),
+                    stream=True
                 ),
                 ensure_ascii=False
             ) + "\n\n"
             yield "data: [DONE]\n\n"
-    
-    else:  # Non-streaming mode
+
+    else:
         try:
-            all_answer_content = ""
-            for answer in canvas.run(stream=False, bypass_begin=True):
-                if answer.get("running_status"):
+            all_content = ""
+            for ans in completion(
+                tenant_id=tenant_id,
+                agent_id=agent_id,
+                session_id=session_id,
+                query=question,
+                user_id=user_id,
+                **kwargs
+            ):
+                if isinstance(ans, str):
+                    ans = json.loads(ans[5:])
+                if ans.get("event") != "message" or not ans.get("data", {}).get("reference", None):
                     continue
-                
-                final_ans["content"] = "\n".join(answer["content"]) if "content" in answer else ""
-                final_ans["reference"] = answer.get("reference", [])
-                all_answer_content += final_ans["content"]
-            
-            final_ans["content"] = all_answer_content
-            
-            # Update conversation
-            canvas.messages.append({"role": "assistant", "content": final_ans["content"], "created_at": time.time(), "id": message_id})
-            canvas.history.append(("assistant", final_ans["content"]))
-            if final_ans.get("reference"):
-                canvas.reference.append(final_ans["reference"])
-            conv.dsl = json.loads(str(canvas))
-            API4ConversationService.append_message(conv.id, conv.to_dict())
-            
-            # Return the response in OpenAI format
+                all_content += ans["data"]["content"]
+
+            completion_tokens = len(tiktokenenc.encode(all_content))
+
             yield get_data_openai(
-                id=session_id,
+                id=session_id or str(uuid4()),
                 model=agent_id,
-                content=final_ans["content"],
-                finish_reason="stop",
-                completion_tokens=len(tiktokenenc.encode(final_ans["content"])),
                 prompt_tokens=prompt_tokens,
-                param=canvas.get_preset_param()  # Added param info like in completion
-            )
-            
-        except Exception as e:
-            traceback.print_exc()
-            conv.dsl = json.loads(str(canvas))
-            API4ConversationService.append_message(conv.id, conv.to_dict())
-            yield get_data_openai(
-                id=session_id,
-                model=agent_id,
-                content="**ERROR**: " + str(e),
+                completion_tokens=completion_tokens,
+                content=all_content,
                 finish_reason="stop",
-                completion_tokens=len(tiktokenenc.encode("**ERROR**: " + str(e))),
-                prompt_tokens=prompt_tokens
+                param=None
             )
 
+        except Exception as e:
+            yield get_data_openai(
+                id=session_id or str(uuid4()),
+                model=agent_id,
+                prompt_tokens=prompt_tokens,
+                completion_tokens=len(tiktokenenc.encode(f"**ERROR**: {str(e)}")),
+                content=f"**ERROR**: {str(e)}",
+                finish_reason="stop",
+                param=None
+            )

api/db/services/dialog_service.py

@@ -22,21 +22,27 @@ from datetime import datetime
 from functools import partial
 from timeit import default_timer as timer
 
+import trio
 from langfuse import Langfuse
+from peewee import fn
 
 from agentic_reasoning import DeepResearcher
 from api import settings
 from api.db import LLMType, ParserType, StatusEnum
 from api.db.db_models import DB, Dialog
 from api.db.services.common_service import CommonService
+from api.db.services.document_service import DocumentService
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.langfuse_service import TenantLangfuseService
-from api.db.services.llm_service import LLMBundle, TenantLLMService
+from api.db.services.llm_service import LLMBundle
+from api.db.services.tenant_llm_service import TenantLLMService
 from api.utils import current_timestamp, datetime_format
+from graphrag.general.mind_map_extractor import MindMapExtractor
 from rag.app.resume import forbidden_select_fields4resume
 from rag.app.tag import label_question
 from rag.nlp.search import index_name
 from rag.prompts import chunks_format, citation_prompt, cross_languages, full_question, kb_prompt, keyword_extraction, message_fit_in
+from rag.prompts.prompts import gen_meta_filter, PROMPT_JINJA_ENV, ASK_SUMMARY
 from rag.utils import num_tokens_from_string, rmSpace
 from rag.utils.tavily_conn import Tavily
 
@@ -95,6 +101,66 @@ class DialogService(CommonService):
 
         return list(chats.dicts())
 
+    @classmethod
+    @DB.connection_context()
+    def get_by_tenant_ids(cls, joined_tenant_ids, user_id, page_number, items_per_page, orderby, desc, keywords, parser_id=None):
+        from api.db.db_models import User
+
+        fields = [
+            cls.model.id,
+            cls.model.tenant_id,
+            cls.model.name,
+            cls.model.description,
+            cls.model.language,
+            cls.model.llm_id,
+            cls.model.llm_setting,
+            cls.model.prompt_type,
+            cls.model.prompt_config,
+            cls.model.similarity_threshold,
+            cls.model.vector_similarity_weight,
+            cls.model.top_n,
+            cls.model.top_k,
+            cls.model.do_refer,
+            cls.model.rerank_id,
+            cls.model.kb_ids,
+            cls.model.icon,
+            cls.model.status,
+            User.nickname,
+            User.avatar.alias("tenant_avatar"),
+            cls.model.update_time,
+            cls.model.create_time,
+        ]
+        if keywords:
+            dialogs = (
+                cls.model.select(*fields)
+                .join(User, on=(cls.model.tenant_id == User.id))
+                .where(
+                    (cls.model.tenant_id.in_(joined_tenant_ids) | (cls.model.tenant_id == user_id)) & (cls.model.status == StatusEnum.VALID.value),
+                    (fn.LOWER(cls.model.name).contains(keywords.lower())),
+                )
+            )
+        else:
+            dialogs = (
+                cls.model.select(*fields)
+                .join(User, on=(cls.model.tenant_id == User.id))
+                .where(
+                    (cls.model.tenant_id.in_(joined_tenant_ids) | (cls.model.tenant_id == user_id)) & (cls.model.status == StatusEnum.VALID.value),
+                )
+            )
+        if parser_id:
+            dialogs = dialogs.where(cls.model.parser_id == parser_id)
+        if desc:
+            dialogs = dialogs.order_by(cls.model.getter_by(orderby).desc())
+        else:
+            dialogs = dialogs.order_by(cls.model.getter_by(orderby).asc())
+
+        count = dialogs.count()
+
+        if page_number and items_per_page:
+            dialogs = dialogs.paginate(page_number, items_per_page)
+
+        return list(dialogs.dicts()), count
+
 
 def chat_solo(dialog, messages, stream=True):
     if TenantLLMService.llm_id2llm_type(dialog.llm_id) == "image2text":
@@ -189,6 +255,55 @@ def repair_bad_citation_formats(answer: str, kbinfos: dict, idx: set):
     return answer, idx
 
 
+def meta_filter(metas: dict, filters: list[dict]):
+    doc_ids = set([])
+
+    def filter_out(v2docs, operator, value):
+        ids = []
+        for input, docids in v2docs.items():
+            try:
+                input = float(input)
+                value = float(value)
+            except Exception:
+                input = str(input)
+                value = str(value)
+
+            for conds in [
+                    (operator == "contains", str(value).lower() in str(input).lower()),
+                    (operator == "not contains", str(value).lower() not in str(input).lower()),
+                    (operator == "start with", str(input).lower().startswith(str(value).lower())),
+                    (operator == "end with", str(input).lower().endswith(str(value).lower())),
+                    (operator == "empty", not input),
+                    (operator == "not empty", input),
+                    (operator == "=", input == value),
+                    (operator == "≠", input != value),
+                    (operator == ">", input > value),
+                    (operator == "<", input < value),
+                    (operator == "≥", input >= value),
+                    (operator == "≤", input <= value),
+                ]:
+                try:
+                    if all(conds):
+                        ids.extend(docids)
+                        break
+                except Exception:
+                    pass
+        return ids
+
+    for k, v2docs in metas.items():
+        for f in filters:
+            if k != f["key"]:
+                continue
+            ids = filter_out(v2docs, f["op"], f["value"])
+            if not doc_ids:
+                doc_ids = set(ids)
+            else:
+                doc_ids = doc_ids & set(ids)
+            if not doc_ids:
+                return []
+    return list(doc_ids)
+
+
 def chat(dialog, messages, stream=True, **kwargs):
     assert messages[-1]["role"] == "user", "The last content of this conversation is not from user."
     if not dialog.kb_ids and not dialog.prompt_config.get("tavily_api_key"):
@@ -208,12 +323,14 @@ def chat(dialog, messages, stream=True, **kwargs):
     check_llm_ts = timer()
 
     langfuse_tracer = None
+    trace_context = {}
     langfuse_keys = TenantLangfuseService.filter_by_tenant(tenant_id=dialog.tenant_id)
     if langfuse_keys:
         langfuse = Langfuse(public_key=langfuse_keys.public_key, secret_key=langfuse_keys.secret_key, host=langfuse_keys.host)
         if langfuse.auth_check():
             langfuse_tracer = langfuse
-            langfuse.trace = langfuse_tracer.trace(name=f"{dialog.name}-{llm_model_config['llm_name']}")
+            trace_id = langfuse_tracer.create_trace_id()
+            trace_context = {"trace_id": trace_id}
 
     check_langfuse_tracer_ts = timer()
     kbs, embd_mdl, rerank_mdl, chat_mdl, tts_mdl = get_models(dialog)
@@ -224,9 +341,10 @@ def chat(dialog, messages, stream=True, **kwargs):
 
     retriever = settings.retrievaler
     questions = [m["content"] for m in messages if m["role"] == "user"][-3:]
-    attachments = kwargs["doc_ids"].split(",") if "doc_ids" in kwargs else None
+    attachments = kwargs["doc_ids"].split(",") if "doc_ids" in kwargs else []
     if "doc_ids" in messages[-1]:
         attachments = messages[-1]["doc_ids"]
+
     prompt_config = dialog.prompt_config
     field_map = KnowledgebaseService.get_field_map(dialog.kb_ids)
     # try to use sql if field mapping is good to go
@@ -253,6 +371,18 @@ def chat(dialog, messages, stream=True, **kwargs):
     if prompt_config.get("cross_languages"):
         questions = [cross_languages(dialog.tenant_id, dialog.llm_id, questions[0], prompt_config["cross_languages"])]
 
+    if dialog.meta_data_filter:
+        metas = DocumentService.get_meta_by_kbs(dialog.kb_ids)
+        if dialog.meta_data_filter.get("method") == "auto":
+            filters = gen_meta_filter(chat_mdl, metas, questions[-1])
+            attachments.extend(meta_filter(metas, filters))
+            if not attachments:
+                attachments = None
+        elif dialog.meta_data_filter.get("method") == "manual":
+            attachments.extend(meta_filter(metas, dialog.meta_data_filter["manual"]))
+            if not attachments:
+                attachments = None
+
     if prompt_config.get("keyword", False):
         questions[-1] += keyword_extraction(chat_mdl, questions[-1])
 
@@ -260,17 +390,26 @@ def chat(dialog, messages, stream=True, **kwargs):
 
     thought = ""
     kbinfos = {"total": 0, "chunks": [], "doc_aggs": []}
+    knowledges = []
 
-    if "knowledge" not in [p["key"] for p in prompt_config["parameters"]]:
-        knowledges = []
-    else:
+    if attachments is not None and "knowledge" in [p["key"] for p in prompt_config["parameters"]]:
         tenant_ids = list(set([kb.tenant_id for kb in kbs]))
         knowledges = []
         if prompt_config.get("reasoning", False):
             reasoner = DeepResearcher(
                 chat_mdl,
                 prompt_config,
-                partial(retriever.retrieval, embd_mdl=embd_mdl, tenant_ids=tenant_ids, kb_ids=dialog.kb_ids, page=1, page_size=dialog.top_n, similarity_threshold=0.2, vector_similarity_weight=0.3),
+                partial(
+                    retriever.retrieval,
+                    embd_mdl=embd_mdl,
+                    tenant_ids=tenant_ids,
+                    kb_ids=dialog.kb_ids,
+                    page=1,
+                    page_size=dialog.top_n,
+                    similarity_threshold=0.2,
+                    vector_similarity_weight=0.3,
+                    doc_ids=attachments,
+                ),
             )
 
             for think in reasoner.thinking(kbinfos, " ".join(questions)):
@@ -400,17 +539,19 @@ def chat(dialog, messages, stream=True, **kwargs):
             f"  - Token speed: {int(tk_num / (generate_result_time_cost / 1000.0))}/s"
         )
 
-        langfuse_output = "\n" + re.sub(r"^.*?(### Query:.*)", r"\1", prompt, flags=re.DOTALL)
-        langfuse_output = {"time_elapsed:": re.sub(r"\n", "  \n", langfuse_output), "created_at": time.time()}
-
         # Add a condition check to call the end method only if langfuse_tracer exists
         if langfuse_tracer and "langfuse_generation" in locals():
-            langfuse_generation.end(output=langfuse_output)
+            langfuse_output = "\n" + re.sub(r"^.*?(### Query:.*)", r"\1", prompt, flags=re.DOTALL)
+            langfuse_output = {"time_elapsed:": re.sub(r"\n", "  \n", langfuse_output), "created_at": time.time()}
+            langfuse_generation.update(output=langfuse_output)
+            langfuse_generation.end()
 
         return {"answer": think + answer, "reference": refs, "prompt": re.sub(r"\n", "  \n", prompt), "created_at": time.time()}
 
     if langfuse_tracer:
-        langfuse_generation = langfuse_tracer.trace.generation(name="chat", model=llm_model_config["llm_name"], input={"prompt": prompt, "prompt4citation": prompt4citation, "messages": msg})
+        langfuse_generation = langfuse_tracer.start_generation(
+            trace_context=trace_context, name="chat", model=llm_model_config["llm_name"], input={"prompt": prompt, "prompt4citation": prompt4citation, "messages": msg}
+        )
 
     if stream:
         last_ans = ""
@@ -556,7 +697,14 @@ def tts(tts_mdl, text):
     return binascii.hexlify(bin).decode("utf-8")
 
 
-def ask(question, kb_ids, tenant_id, chat_llm_name=None):
+def ask(question, kb_ids, tenant_id, chat_llm_name=None, search_config={}):
+    doc_ids = search_config.get("doc_ids", [])
+    rerank_mdl = None
+    kb_ids = search_config.get("kb_ids", kb_ids)
+    chat_llm_name = search_config.get("chat_id", chat_llm_name)
+    rerank_id = search_config.get("rerank_id", "")
+    meta_data_filter = search_config.get("meta_data_filter")
+
     kbs = KnowledgebaseService.get_by_ids(kb_ids)
     embedding_list = list(set([kb.embd_id for kb in kbs]))
 
@@ -565,30 +713,46 @@ def ask(question, kb_ids, tenant_id, chat_llm_name=None):
 
     embd_mdl = LLMBundle(tenant_id, LLMType.EMBEDDING, embedding_list[0])
     chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, chat_llm_name)
+    if rerank_id:
+        rerank_mdl = LLMBundle(tenant_id, LLMType.RERANK, rerank_id)
     max_tokens = chat_mdl.max_length
     tenant_ids = list(set([kb.tenant_id for kb in kbs]))
-    kbinfos = retriever.retrieval(question, embd_mdl, tenant_ids, kb_ids, 1, 12, 0.1, 0.3, aggs=False, rank_feature=label_question(question, kbs))
+
+    if meta_data_filter:
+        metas = DocumentService.get_meta_by_kbs(kb_ids)
+        if meta_data_filter.get("method") == "auto":
+            filters = gen_meta_filter(chat_mdl, metas, question)
+            doc_ids.extend(meta_filter(metas, filters))
+            if not doc_ids:
+                doc_ids = None
+        elif meta_data_filter.get("method") == "manual":
+            doc_ids.extend(meta_filter(metas, meta_data_filter["manual"]))
+            if not doc_ids:
+                doc_ids = None
+
+    kbinfos = retriever.retrieval(
+        question = question,
+        embd_mdl=embd_mdl,
+        tenant_ids=tenant_ids,
+        kb_ids=kb_ids,
+        page=1,
+        page_size=12,
+        similarity_threshold=search_config.get("similarity_threshold", 0.1),
+        vector_similarity_weight=search_config.get("vector_similarity_weight", 0.3),
+        top=search_config.get("top_k", 1024),
+        doc_ids=doc_ids,
+        aggs=False,
+        rerank_mdl=rerank_mdl,
+        rank_feature=label_question(question, kbs)
+    )
+
     knowledges = kb_prompt(kbinfos, max_tokens)
-    prompt = """
-    Role: You're a smart assistant. Your name is Miss R.
-    Task: Summarize the information from knowledge bases and answer user's question.
-    Requirements and restriction:
-      - DO NOT make things up, especially for numbers.
-      - If the information from knowledge is irrelevant with user's question, JUST SAY: Sorry, no relevant information provided.
-      - Answer with markdown format text.
-      - Answer in language of user's question.
-      - DO NOT make things up, especially for numbers.
+    sys_prompt = PROMPT_JINJA_ENV.from_string(ASK_SUMMARY).render(knowledge="\n".join(knowledges))
 
-    ### Information from knowledge bases
-    %s
-
-    The above is information from knowledge bases.
-
-    """ % "\n".join(knowledges)
     msg = [{"role": "user", "content": question}]
 
     def decorate_answer(answer):
-        nonlocal knowledges, kbinfos, prompt
+        nonlocal knowledges, kbinfos, sys_prompt
         answer, idx = retriever.insert_citations(answer, [ck["content_ltks"] for ck in kbinfos["chunks"]], [ck["vector"] for ck in kbinfos["chunks"]], embd_mdl, tkweight=0.7, vtweight=0.3)
         idx = set([kbinfos["chunks"][int(i)]["doc_id"] for i in idx])
         recall_docs = [d for d in kbinfos["doc_aggs"] if d["doc_id"] in idx]
@@ -606,7 +770,55 @@ def ask(question, kb_ids, tenant_id, chat_llm_name=None):
         return {"answer": answer, "reference": refs}
 
     answer = ""
-    for ans in chat_mdl.chat_streamly(prompt, msg, {"temperature": 0.1}):
+    for ans in chat_mdl.chat_streamly(sys_prompt, msg, {"temperature": 0.1}):
         answer = ans
         yield {"answer": answer, "reference": {}}
     yield decorate_answer(answer)
+
+
+def gen_mindmap(question, kb_ids, tenant_id, search_config={}):
+    meta_data_filter = search_config.get("meta_data_filter", {})
+    doc_ids = search_config.get("doc_ids", [])
+    rerank_id = search_config.get("rerank_id", "")
+    rerank_mdl = None
+    kbs = KnowledgebaseService.get_by_ids(kb_ids)
+    if not kbs:
+        return {"error": "No KB selected"}
+    embedding_list = list(set([kb.embd_id for kb in kbs]))
+    tenant_ids = list(set([kb.tenant_id for kb in kbs]))
+
+    embd_mdl = LLMBundle(tenant_id, LLMType.EMBEDDING, llm_name=embedding_list[0])
+    chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, llm_name=search_config.get("chat_id", ""))
+    if rerank_id:
+        rerank_mdl = LLMBundle(tenant_id, LLMType.RERANK, rerank_id)
+
+    if meta_data_filter:
+        metas = DocumentService.get_meta_by_kbs(kb_ids)
+        if meta_data_filter.get("method") == "auto":
+            filters = gen_meta_filter(chat_mdl, metas, question)
+            doc_ids.extend(meta_filter(metas, filters))
+            if not doc_ids:
+                doc_ids = None
+        elif meta_data_filter.get("method") == "manual":
+            doc_ids.extend(meta_filter(metas, meta_data_filter["manual"]))
+            if not doc_ids:
+                doc_ids = None
+
+    ranks = settings.retrievaler.retrieval(
+        question=question,
+        embd_mdl=embd_mdl,
+        tenant_ids=tenant_ids,
+        kb_ids=kb_ids,
+        page=1,
+        page_size=12,
+        similarity_threshold=search_config.get("similarity_threshold", 0.2),
+        vector_similarity_weight=search_config.get("vector_similarity_weight", 0.3),
+        top=search_config.get("top_k", 1024),
+        doc_ids=doc_ids,
+        aggs=False,
+        rerank_mdl=rerank_mdl,
+        rank_feature=label_question(question, kbs),
+    )
+    mindmap = MindMapExtractor(chat_mdl)
+    mind_map = trio.run(mindmap, [c["content_with_weight"] for c in ranks["chunks"]])
+    return mind_map.output

api/db/services/document_service.py

@@ -243,7 +243,7 @@ class DocumentService(CommonService):
         from api.db.services.task_service import TaskService
         cls.clear_chunk_num(doc.id)
         try:
-            TaskService.filter_delete(Task.doc_id == doc.id)
+            TaskService.filter_delete([Task.doc_id == doc.id])
             page = 0
             page_size = 1000
             all_chunk_ids = []
@@ -574,6 +574,25 @@ class DocumentService(CommonService):
     def update_meta_fields(cls, doc_id, meta_fields):
         return cls.update_by_id(doc_id, {"meta_fields": meta_fields})
 
+    @classmethod
+    @DB.connection_context()
+    def get_meta_by_kbs(cls, kb_ids):
+        fields = [
+            cls.model.id,
+            cls.model.meta_fields,
+        ]
+        meta = {}
+        for r in cls.model.select(*fields).where(cls.model.kb_id.in_(kb_ids)):
+            doc_id = r.id
+            for k,v in r.meta_fields.items():
+                if k not in meta:
+                    meta[k] = {}
+                v = str(v)
+                if v not in meta[k]:
+                    meta[k][v] = []
+                meta[k][v].append(doc_id)
+        return meta
+
     @classmethod
     @DB.connection_context()
     def update_progress(cls):

api/db/services/file_service.py

@@ -227,10 +227,13 @@ class FileService(CommonService):
         #     tenant_id: Tenant ID
         # Returns:
         #     Knowledge base folder dictionary
-        for root in cls.model.select().where((cls.model.tenant_id == tenant_id), (cls.model.parent_id == cls.model.id)):
-            for folder in cls.model.select().where((cls.model.tenant_id == tenant_id), (cls.model.parent_id == root.id), (cls.model.name == KNOWLEDGEBASE_FOLDER_NAME)):
-                return folder.to_dict()
-        assert False, "Can't find the KB folder. Database init error."
+        root_folder = cls.get_root_folder(tenant_id)
+        root_id = root_folder["id"]
+        kb_folder = cls.model.select().where((cls.model.tenant_id == tenant_id), (cls.model.parent_id == root_id), (cls.model.name == KNOWLEDGEBASE_FOLDER_NAME)).first()
+        if not kb_folder:
+            kb_folder = cls.new_a_file_from_kb(tenant_id, KNOWLEDGEBASE_FOLDER_NAME, root_id)
+            return kb_folder
+        return kb_folder.to_dict()
 
     @classmethod
     @DB.connection_context()
@@ -499,10 +502,9 @@ class FileService(CommonService):
     @staticmethod
     def get_blob(user_id, location):
         bname = f"{user_id}-downloads"
-        return  STORAGE_IMPL.get(bname, location)
+        return STORAGE_IMPL.get(bname, location)
 
     @staticmethod
     def put_blob(user_id, location, blob):
         bname = f"{user_id}-downloads"
-        return  STORAGE_IMPL.put(bname, location, blob)
-
+        return STORAGE_IMPL.put(bname, location, blob)

api/db/services/llm_service.py

@@ -13,240 +13,78 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 #
+import inspect
 import logging
 import re
 from functools import partial
 from typing import Generator
-
-from langfuse import Langfuse
-
-from api import settings
-from api.db import LLMType
-from api.db.db_models import DB, LLM, LLMFactories, TenantLLM
+from api.db.db_models import LLM
 from api.db.services.common_service import CommonService
-from api.db.services.langfuse_service import TenantLangfuseService
-from api.db.services.user_service import TenantService
-from rag.llm import ChatModel, CvModel, EmbeddingModel, RerankModel, Seq2txtModel, TTSModel
-
-
-class LLMFactoriesService(CommonService):
-    model = LLMFactories
+from api.db.services.tenant_llm_service import LLM4Tenant, TenantLLMService
 
 
 class LLMService(CommonService):
     model = LLM
 
 
-class TenantLLMService(CommonService):
-    model = TenantLLM
+def get_init_tenant_llm(user_id):
+    from api import settings
+    tenant_llm = []
 
-    @classmethod
-    @DB.connection_context()
-    def get_api_key(cls, tenant_id, model_name):
-        mdlnm, fid = TenantLLMService.split_model_name_and_factory(model_name)
-        if not fid:
-            objs = cls.query(tenant_id=tenant_id, llm_name=mdlnm)
-        else:
-            objs = cls.query(tenant_id=tenant_id, llm_name=mdlnm, llm_factory=fid)
+    seen = set()
+    factory_configs = []
+    for factory_config in [
+        settings.CHAT_CFG,
+        settings.EMBEDDING_CFG,
+        settings.ASR_CFG,
+        settings.IMAGE2TEXT_CFG,
+        settings.RERANK_CFG,
+    ]:
+        factory_name = factory_config["factory"]
+        if factory_name not in seen:
+            seen.add(factory_name)
+            factory_configs.append(factory_config)
 
-        if (not objs) and fid:
-            if fid == "LocalAI":
-                mdlnm += "___LocalAI"
-            elif fid == "HuggingFace":
-                mdlnm += "___HuggingFace"
-            elif fid == "OpenAI-API-Compatible":
-                mdlnm += "___OpenAI-API"
-            elif fid == "VLLM":
-                mdlnm += "___VLLM"
-            objs = cls.query(tenant_id=tenant_id, llm_name=mdlnm, llm_factory=fid)
-        if not objs:
-            return
-        return objs[0]
-
-    @classmethod
-    @DB.connection_context()
-    def get_my_llms(cls, tenant_id):
-        fields = [cls.model.llm_factory, LLMFactories.logo, LLMFactories.tags, cls.model.model_type, cls.model.llm_name, cls.model.used_tokens]
-        objs = cls.model.select(*fields).join(LLMFactories, on=(cls.model.llm_factory == LLMFactories.name)).where(cls.model.tenant_id == tenant_id, ~cls.model.api_key.is_null()).dicts()
-
-        return list(objs)
-
-    @staticmethod
-    def split_model_name_and_factory(model_name):
-        arr = model_name.split("@")
-        if len(arr) < 2:
-            return model_name, None
-        if len(arr) > 2:
-            return "@".join(arr[0:-1]), arr[-1]
-
-        # model name must be xxx@yyy
-        try:
-            model_factories = settings.FACTORY_LLM_INFOS
-            model_providers = set([f["name"] for f in model_factories])
-            if arr[-1] not in model_providers:
-                return model_name, None
-            return arr[0], arr[-1]
-        except Exception as e:
-            logging.exception(f"TenantLLMService.split_model_name_and_factory got exception: {e}")
-        return model_name, None
-
-    @classmethod
-    @DB.connection_context()
-    def get_model_config(cls, tenant_id, llm_type, llm_name=None):
-        e, tenant = TenantService.get_by_id(tenant_id)
-        if not e:
-            raise LookupError("Tenant not found")
-
-        if llm_type == LLMType.EMBEDDING.value:
-            mdlnm = tenant.embd_id if not llm_name else llm_name
-        elif llm_type == LLMType.SPEECH2TEXT.value:
-            mdlnm = tenant.asr_id
-        elif llm_type == LLMType.IMAGE2TEXT.value:
-            mdlnm = tenant.img2txt_id if not llm_name else llm_name
-        elif llm_type == LLMType.CHAT.value:
-            mdlnm = tenant.llm_id if not llm_name else llm_name
-        elif llm_type == LLMType.RERANK:
-            mdlnm = tenant.rerank_id if not llm_name else llm_name
-        elif llm_type == LLMType.TTS:
-            mdlnm = tenant.tts_id if not llm_name else llm_name
-        else:
-            assert False, "LLM type error"
-
-        model_config = cls.get_api_key(tenant_id, mdlnm)
-        mdlnm, fid = TenantLLMService.split_model_name_and_factory(mdlnm)
-        if not model_config:  # for some cases seems fid mismatch
-            model_config = cls.get_api_key(tenant_id, mdlnm)
-        if model_config:
-            model_config = model_config.to_dict()
-            llm = LLMService.query(llm_name=mdlnm) if not fid else LLMService.query(llm_name=mdlnm, fid=fid)
-            if not llm and fid:  # for some cases seems fid mismatch
-                llm = LLMService.query(llm_name=mdlnm)
-            if llm:
-                model_config["is_tools"] = llm[0].is_tools
-        if not model_config:
-            if llm_type in [LLMType.EMBEDDING, LLMType.RERANK]:
-                llm = LLMService.query(llm_name=mdlnm) if not fid else LLMService.query(llm_name=mdlnm, fid=fid)
-                if llm and llm[0].fid in ["Youdao", "FastEmbed", "BAAI"]:
-                    model_config = {"llm_factory": llm[0].fid, "api_key": "", "llm_name": mdlnm, "api_base": ""}
-            if not model_config:
-                if mdlnm == "flag-embedding":
-                    model_config = {"llm_factory": "Tongyi-Qianwen", "api_key": "", "llm_name": llm_name, "api_base": ""}
-                else:
-                    if not mdlnm:
-                        raise LookupError(f"Type of {llm_type} model is not set.")
-                    raise LookupError("Model({}) not authorized".format(mdlnm))
-        return model_config
-
-    @classmethod
-    @DB.connection_context()
-    def model_instance(cls, tenant_id, llm_type, llm_name=None, lang="Chinese", **kwargs):
-        model_config = TenantLLMService.get_model_config(tenant_id, llm_type, llm_name)
-        if llm_type == LLMType.EMBEDDING.value:
-            if model_config["llm_factory"] not in EmbeddingModel:
-                return
-            return EmbeddingModel[model_config["llm_factory"]](model_config["api_key"], model_config["llm_name"], base_url=model_config["api_base"])
-
-        if llm_type == LLMType.RERANK:
-            if model_config["llm_factory"] not in RerankModel:
-                return
-            return RerankModel[model_config["llm_factory"]](model_config["api_key"], model_config["llm_name"], base_url=model_config["api_base"])
-
-        if llm_type == LLMType.IMAGE2TEXT.value:
-            if model_config["llm_factory"] not in CvModel:
-                return
-            return CvModel[model_config["llm_factory"]](model_config["api_key"], model_config["llm_name"], lang, base_url=model_config["api_base"], **kwargs)
-
-        if llm_type == LLMType.CHAT.value:
-            if model_config["llm_factory"] not in ChatModel:
-                return
-            return ChatModel[model_config["llm_factory"]](model_config["api_key"], model_config["llm_name"], base_url=model_config["api_base"], **kwargs)
-
-        if llm_type == LLMType.SPEECH2TEXT:
-            if model_config["llm_factory"] not in Seq2txtModel:
-                return
-            return Seq2txtModel[model_config["llm_factory"]](key=model_config["api_key"], model_name=model_config["llm_name"], lang=lang, base_url=model_config["api_base"])
-        if llm_type == LLMType.TTS:
-            if model_config["llm_factory"] not in TTSModel:
-                return
-            return TTSModel[model_config["llm_factory"]](
-                model_config["api_key"],
-                model_config["llm_name"],
-                base_url=model_config["api_base"],
+    for factory_config in factory_configs:
+        for llm in LLMService.query(fid=factory_config["factory"]):
+            tenant_llm.append(
+                {
+                    "tenant_id": user_id,
+                    "llm_factory": factory_config["factory"],
+                    "llm_name": llm.llm_name,
+                    "model_type": llm.model_type,
+                    "api_key": factory_config["api_key"],
+                    "api_base": factory_config["base_url"],
+                    "max_tokens": llm.max_tokens if llm.max_tokens else 8192,
+                }
             )
 
-    @classmethod
-    @DB.connection_context()
-    def increase_usage(cls, tenant_id, llm_type, used_tokens, llm_name=None):
-        e, tenant = TenantService.get_by_id(tenant_id)
-        if not e:
-            logging.error(f"Tenant not found: {tenant_id}")
-            return 0
-
-        llm_map = {
-            LLMType.EMBEDDING.value: tenant.embd_id if not llm_name else llm_name,
-            LLMType.SPEECH2TEXT.value: tenant.asr_id,
-            LLMType.IMAGE2TEXT.value: tenant.img2txt_id,
-            LLMType.CHAT.value: tenant.llm_id if not llm_name else llm_name,
-            LLMType.RERANK.value: tenant.rerank_id if not llm_name else llm_name,
-            LLMType.TTS.value: tenant.tts_id if not llm_name else llm_name,
-        }
-
-        mdlnm = llm_map.get(llm_type)
-        if mdlnm is None:
-            logging.error(f"LLM type error: {llm_type}")
-            return 0
-
-        llm_name, llm_factory = TenantLLMService.split_model_name_and_factory(mdlnm)
-
-        try:
-            num = (
-                cls.model.update(used_tokens=cls.model.used_tokens + used_tokens)
-                .where(cls.model.tenant_id == tenant_id, cls.model.llm_name == llm_name, cls.model.llm_factory == llm_factory if llm_factory else True)
-                .execute()
+    if settings.LIGHTEN != 1:
+        for buildin_embedding_model in settings.BUILTIN_EMBEDDING_MODELS:
+            mdlnm, fid = TenantLLMService.split_model_name_and_factory(buildin_embedding_model)
+            tenant_llm.append(
+                {
+                    "tenant_id": user_id,
+                    "llm_factory": fid,
+                    "llm_name": mdlnm,
+                    "model_type": "embedding",
+                    "api_key": "",
+                    "api_base": "",
+                    "max_tokens": 1024 if buildin_embedding_model == "BAAI/bge-large-zh-v1.5@BAAI" else 512,
+                }
             )
-        except Exception:
-            logging.exception("TenantLLMService.increase_usage got exception,Failed to update used_tokens for tenant_id=%s, llm_name=%s", tenant_id, llm_name)
-            return 0
 
-        return num
-
-    @classmethod
-    @DB.connection_context()
-    def get_openai_models(cls):
-        objs = cls.model.select().where((cls.model.llm_factory == "OpenAI"), ~(cls.model.llm_name == "text-embedding-3-small"), ~(cls.model.llm_name == "text-embedding-3-large")).dicts()
-        return list(objs)
-
-    @staticmethod
-    def llm_id2llm_type(llm_id: str) ->str|None:
-        llm_id, *_ = TenantLLMService.split_model_name_and_factory(llm_id)
-        llm_factories = settings.FACTORY_LLM_INFOS
-        for llm_factory in llm_factories:
-            for llm in llm_factory["llm"]:
-                if llm_id == llm["llm_name"]:
-                    return llm["model_type"].split(",")[-1]
+    unique = {}
+    for item in tenant_llm:
+        key = (item["tenant_id"], item["llm_factory"], item["llm_name"])
+        if key not in unique:
+            unique[key] = item
+    return list(unique.values())
 
 
-class LLMBundle:
+class LLMBundle(LLM4Tenant):
     def __init__(self, tenant_id, llm_type, llm_name=None, lang="Chinese", **kwargs):
-        self.tenant_id = tenant_id
-        self.llm_type = llm_type
-        self.llm_name = llm_name
-        self.mdl = TenantLLMService.model_instance(tenant_id, llm_type, llm_name, lang=lang, **kwargs)
-        assert self.mdl, "Can't find model for {}/{}/{}".format(tenant_id, llm_type, llm_name)
-        model_config = TenantLLMService.get_model_config(tenant_id, llm_type, llm_name)
-        self.max_length = model_config.get("max_tokens", 8192)
-
-        self.is_tools = model_config.get("is_tools", False)
-        self.verbose_tool_use = kwargs.get("verbose_tool_use")
-
-        langfuse_keys = TenantLangfuseService.filter_by_tenant(tenant_id=tenant_id)
-        if langfuse_keys:
-            langfuse = Langfuse(public_key=langfuse_keys.public_key, secret_key=langfuse_keys.secret_key, host=langfuse_keys.host)
-            if langfuse.auth_check():
-                self.langfuse = langfuse
-                self.trace = self.langfuse.trace(name=f"{self.llm_type}-{self.llm_name}")
-        else:
-            self.langfuse = None
+        super().__init__(tenant_id, llm_type, llm_name, lang, **kwargs)
 
     def bind_tools(self, toolcall_session, tools):
         if not self.is_tools:
@@ -256,7 +94,7 @@ class LLMBundle:
 
     def encode(self, texts: list):
         if self.langfuse:
-            generation = self.trace.generation(name="encode", model=self.llm_name, input={"texts": texts})
+            generation = self.langfuse.start_generation(trace_context=self.trace_context, name="encode", model=self.llm_name, input={"texts": texts})
 
         embeddings, used_tokens = self.mdl.encode(texts)
         llm_name = getattr(self, "llm_name", None)
@@ -264,13 +102,14 @@ class LLMBundle:
             logging.error("LLMBundle.encode can't update token usage for {}/EMBEDDING used_tokens: {}".format(self.tenant_id, used_tokens))
 
         if self.langfuse:
-            generation.end(usage_details={"total_tokens": used_tokens})
+            generation.update(usage_details={"total_tokens": used_tokens})
+            generation.end()
 
         return embeddings, used_tokens
 
     def encode_queries(self, query: str):
         if self.langfuse:
-            generation = self.trace.generation(name="encode_queries", model=self.llm_name, input={"query": query})
+            generation = self.langfuse.start_generation(trace_context=self.trace_context, name="encode_queries", model=self.llm_name, input={"query": query})
 
         emd, used_tokens = self.mdl.encode_queries(query)
         llm_name = getattr(self, "llm_name", None)
@@ -278,65 +117,70 @@ class LLMBundle:
             logging.error("LLMBundle.encode_queries can't update token usage for {}/EMBEDDING used_tokens: {}".format(self.tenant_id, used_tokens))
 
         if self.langfuse:
-            generation.end(usage_details={"total_tokens": used_tokens})
+            generation.update(usage_details={"total_tokens": used_tokens})
+            generation.end()
 
         return emd, used_tokens
 
     def similarity(self, query: str, texts: list):
         if self.langfuse:
-            generation = self.trace.generation(name="similarity", model=self.llm_name, input={"query": query, "texts": texts})
+            generation = self.langfuse.start_generation(trace_context=self.trace_context, name="similarity", model=self.llm_name, input={"query": query, "texts": texts})
 
         sim, used_tokens = self.mdl.similarity(query, texts)
         if not TenantLLMService.increase_usage(self.tenant_id, self.llm_type, used_tokens):
             logging.error("LLMBundle.similarity can't update token usage for {}/RERANK used_tokens: {}".format(self.tenant_id, used_tokens))
 
         if self.langfuse:
-            generation.end(usage_details={"total_tokens": used_tokens})
+            generation.update(usage_details={"total_tokens": used_tokens})
+            generation.end()
 
         return sim, used_tokens
 
     def describe(self, image, max_tokens=300):
         if self.langfuse:
-            generation = self.trace.generation(name="describe", metadata={"model": self.llm_name})
+            generation = self.langfuse.start_generation(trace_context=self.trace_context, name="describe", metadata={"model": self.llm_name})
 
         txt, used_tokens = self.mdl.describe(image)
         if not TenantLLMService.increase_usage(self.tenant_id, self.llm_type, used_tokens):
             logging.error("LLMBundle.describe can't update token usage for {}/IMAGE2TEXT used_tokens: {}".format(self.tenant_id, used_tokens))
 
         if self.langfuse:
-            generation.end(output={"output": txt}, usage_details={"total_tokens": used_tokens})
+            generation.update(output={"output": txt}, usage_details={"total_tokens": used_tokens})
+            generation.end()
 
         return txt
 
     def describe_with_prompt(self, image, prompt):
         if self.langfuse:
-            generation = self.trace.generation(name="describe_with_prompt", metadata={"model": self.llm_name, "prompt": prompt})
+            generation = self.langfuse.start_generation(trace_context=self.trace_context, name="describe_with_prompt", metadata={"model": self.llm_name, "prompt": prompt})
 
         txt, used_tokens = self.mdl.describe_with_prompt(image, prompt)
         if not TenantLLMService.increase_usage(self.tenant_id, self.llm_type, used_tokens):
             logging.error("LLMBundle.describe can't update token usage for {}/IMAGE2TEXT used_tokens: {}".format(self.tenant_id, used_tokens))
 
         if self.langfuse:
-            generation.end(output={"output": txt}, usage_details={"total_tokens": used_tokens})
+            generation.update(output={"output": txt}, usage_details={"total_tokens": used_tokens})
+            generation.end()
 
         return txt
 
     def transcription(self, audio):
         if self.langfuse:
-            generation = self.trace.generation(name="transcription", metadata={"model": self.llm_name})
+            generation = self.langfuse.start_generation(trace_context=self.trace_context, name="transcription", metadata={"model": self.llm_name})
 
         txt, used_tokens = self.mdl.transcription(audio)
         if not TenantLLMService.increase_usage(self.tenant_id, self.llm_type, used_tokens):
             logging.error("LLMBundle.transcription can't update token usage for {}/SEQUENCE2TXT used_tokens: {}".format(self.tenant_id, used_tokens))
 
         if self.langfuse:
-            generation.end(output={"output": txt}, usage_details={"total_tokens": used_tokens})
+            generation.update(output={"output": txt}, usage_details={"total_tokens": used_tokens})
+            generation.end()
 
         return txt
 
     def tts(self, text: str) -> Generator[bytes, None, None]:
         if self.langfuse:
-            span = self.trace.span(name="tts", input={"text": text})
+            generation = self.langfuse.start_generation(trace_context=self.trace_context, name="tts", input={"text": text})
 
         for chunk in self.mdl.tts(text):
             if isinstance(chunk, int):
@@ -346,7 +190,7 @@ class LLMBundle:
             yield chunk
 
         if self.langfuse:
-            span.end()
+            generation.end()
 
     def _remove_reasoning_content(self, txt: str) -> str:
         first_think_start = txt.find("<think>")
@@ -361,16 +205,34 @@ class LLMBundle:
             return txt
 
         return txt[last_think_end + len("</think>") :]
+    
+    @staticmethod
+    def _clean_param(chat_partial, **kwargs):
+        func = chat_partial.func
+        sig = inspect.signature(func)
+        keyword_args = []
+        support_var_args = False
+        for param in sig.parameters.values():
+            if param.kind == inspect.Parameter.VAR_KEYWORD or param.kind == inspect.Parameter.VAR_POSITIONAL:
+                support_var_args = True
+            elif param.kind == inspect.Parameter.KEYWORD_ONLY:
+                keyword_args.append(param.name)
 
-    def chat(self, system: str, history: list, gen_conf: dict={}, **kwargs) -> str:
+        use_kwargs = kwargs
+        if not support_var_args:
+            use_kwargs = {k: v for k, v in kwargs.items() if k in keyword_args}
+        return use_kwargs
+        
+    def chat(self, system: str, history: list, gen_conf: dict = {}, **kwargs) -> str:
         if self.langfuse:
-            generation = self.trace.generation(name="chat", model=self.llm_name, input={"system": system, "history": history})
+            generation = self.langfuse.start_generation(trace_context=self.trace_context, name="chat", model=self.llm_name, input={"system": system, "history": history})
 
         chat_partial = partial(self.mdl.chat, system, history, gen_conf)
         if self.is_tools and self.mdl.is_tools:
             chat_partial = partial(self.mdl.chat_with_tools, system, history, gen_conf)
-
-        txt, used_tokens = chat_partial(**kwargs)
+            
+        use_kwargs = self._clean_param(chat_partial, **kwargs)
+        txt, used_tokens = chat_partial(**use_kwargs)
         txt = self._remove_reasoning_content(txt)
 
         if not self.verbose_tool_use:
@@ -380,25 +242,27 @@ class LLMBundle:
             logging.error("LLMBundle.chat can't update token usage for {}/CHAT llm_name: {}, used_tokens: {}".format(self.tenant_id, self.llm_name, used_tokens))
 
         if self.langfuse:
-            generation.end(output={"output": txt}, usage_details={"total_tokens": used_tokens})
+            generation.update(output={"output": txt}, usage_details={"total_tokens": used_tokens})
+            generation.end()
 
         return txt
 
-    def chat_streamly(self, system: str, history: list, gen_conf: dict={}, **kwargs):
+    def chat_streamly(self, system: str, history: list, gen_conf: dict = {}, **kwargs):
         if self.langfuse:
-            generation = self.trace.generation(name="chat_streamly", model=self.llm_name, input={"system": system, "history": history})
+            generation = self.langfuse.start_generation(trace_context=self.trace_context, name="chat_streamly", model=self.llm_name, input={"system": system, "history": history})
 
         ans = ""
         chat_partial = partial(self.mdl.chat_streamly, system, history, gen_conf)
         total_tokens = 0
         if self.is_tools and self.mdl.is_tools:
             chat_partial = partial(self.mdl.chat_streamly_with_tools, system, history, gen_conf)
-
-        for txt in chat_partial(**kwargs):
+        use_kwargs = self._clean_param(chat_partial, **kwargs)
+        for txt in chat_partial(**use_kwargs):
             if isinstance(txt, int):
                 total_tokens = txt
                 if self.langfuse:
-                    generation.end(output={"output": ans})
+                    generation.update(output={"output": ans})
+                    generation.end()
                 break
 
             if txt.endswith("</think>"):

api/db/services/search_service.py

@@ -71,6 +71,8 @@ class SearchService(CommonService):
             .first()
             .to_dict()
         )
+        if not search:
+            return {}
         return search
 
     @classmethod

api/db/services/tenant_llm_service.py

@@ -0,0 +1,252 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+import logging
+from langfuse import Langfuse
+from api import settings
+from api.db import LLMType
+from api.db.db_models import DB, LLMFactories, TenantLLM
+from api.db.services.common_service import CommonService
+from api.db.services.langfuse_service import TenantLangfuseService
+from api.db.services.user_service import TenantService
+from rag.llm import ChatModel, CvModel, EmbeddingModel, RerankModel, Seq2txtModel, TTSModel
+
+
+class LLMFactoriesService(CommonService):
+    model = LLMFactories
+
+
+class TenantLLMService(CommonService):
+    model = TenantLLM
+
+    @classmethod
+    @DB.connection_context()
+    def get_api_key(cls, tenant_id, model_name):
+        mdlnm, fid = TenantLLMService.split_model_name_and_factory(model_name)
+        if not fid:
+            objs = cls.query(tenant_id=tenant_id, llm_name=mdlnm)
+        else:
+            objs = cls.query(tenant_id=tenant_id, llm_name=mdlnm, llm_factory=fid)
+
+        if (not objs) and fid:
+            if fid == "LocalAI":
+                mdlnm += "___LocalAI"
+            elif fid == "HuggingFace":
+                mdlnm += "___HuggingFace"
+            elif fid == "OpenAI-API-Compatible":
+                mdlnm += "___OpenAI-API"
+            elif fid == "VLLM":
+                mdlnm += "___VLLM"
+            objs = cls.query(tenant_id=tenant_id, llm_name=mdlnm, llm_factory=fid)
+        if not objs:
+            return
+        return objs[0]
+
+    @classmethod
+    @DB.connection_context()
+    def get_my_llms(cls, tenant_id):
+        fields = [cls.model.llm_factory, LLMFactories.logo, LLMFactories.tags, cls.model.model_type, cls.model.llm_name, cls.model.used_tokens]
+        objs = cls.model.select(*fields).join(LLMFactories, on=(cls.model.llm_factory == LLMFactories.name)).where(cls.model.tenant_id == tenant_id, ~cls.model.api_key.is_null()).dicts()
+
+        return list(objs)
+
+    @staticmethod
+    def split_model_name_and_factory(model_name):
+        arr = model_name.split("@")
+        if len(arr) < 2:
+            return model_name, None
+        if len(arr) > 2:
+            return "@".join(arr[0:-1]), arr[-1]
+
+        # model name must be xxx@yyy
+        try:
+            model_factories = settings.FACTORY_LLM_INFOS
+            model_providers = set([f["name"] for f in model_factories])
+            if arr[-1] not in model_providers:
+                return model_name, None
+            return arr[0], arr[-1]
+        except Exception as e:
+            logging.exception(f"TenantLLMService.split_model_name_and_factory got exception: {e}")
+        return model_name, None
+
+    @classmethod
+    @DB.connection_context()
+    def get_model_config(cls, tenant_id, llm_type, llm_name=None):
+        from api.db.services.llm_service import LLMService
+        e, tenant = TenantService.get_by_id(tenant_id)
+        if not e:
+            raise LookupError("Tenant not found")
+
+        if llm_type == LLMType.EMBEDDING.value:
+            mdlnm = tenant.embd_id if not llm_name else llm_name
+        elif llm_type == LLMType.SPEECH2TEXT.value:
+            mdlnm = tenant.asr_id
+        elif llm_type == LLMType.IMAGE2TEXT.value:
+            mdlnm = tenant.img2txt_id if not llm_name else llm_name
+        elif llm_type == LLMType.CHAT.value:
+            mdlnm = tenant.llm_id if not llm_name else llm_name
+        elif llm_type == LLMType.RERANK:
+            mdlnm = tenant.rerank_id if not llm_name else llm_name
+        elif llm_type == LLMType.TTS:
+            mdlnm = tenant.tts_id if not llm_name else llm_name
+        else:
+            assert False, "LLM type error"
+
+        model_config = cls.get_api_key(tenant_id, mdlnm)
+        mdlnm, fid = TenantLLMService.split_model_name_and_factory(mdlnm)
+        if not model_config:  # for some cases seems fid mismatch
+            model_config = cls.get_api_key(tenant_id, mdlnm)
+        if model_config:
+            model_config = model_config.to_dict()
+            llm = LLMService.query(llm_name=mdlnm) if not fid else LLMService.query(llm_name=mdlnm, fid=fid)
+            if not llm and fid:  # for some cases seems fid mismatch
+                llm = LLMService.query(llm_name=mdlnm)
+            if llm:
+                model_config["is_tools"] = llm[0].is_tools
+        if not model_config:
+            if llm_type in [LLMType.EMBEDDING, LLMType.RERANK]:
+                llm = LLMService.query(llm_name=mdlnm) if not fid else LLMService.query(llm_name=mdlnm, fid=fid)
+                if llm and llm[0].fid in ["Youdao", "FastEmbed", "BAAI"]:
+                    model_config = {"llm_factory": llm[0].fid, "api_key": "", "llm_name": mdlnm, "api_base": ""}
+            if not model_config:
+                if mdlnm == "flag-embedding":
+                    model_config = {"llm_factory": "Tongyi-Qianwen", "api_key": "", "llm_name": llm_name, "api_base": ""}
+                else:
+                    if not mdlnm:
+                        raise LookupError(f"Type of {llm_type} model is not set.")
+                    raise LookupError("Model({}) not authorized".format(mdlnm))
+        return model_config
+
+    @classmethod
+    @DB.connection_context()
+    def model_instance(cls, tenant_id, llm_type, llm_name=None, lang="Chinese", **kwargs):
+        model_config = TenantLLMService.get_model_config(tenant_id, llm_type, llm_name)
+        kwargs.update({"provider": model_config["llm_factory"]})
+        if llm_type == LLMType.EMBEDDING.value:
+            if model_config["llm_factory"] not in EmbeddingModel:
+                return
+            return EmbeddingModel[model_config["llm_factory"]](model_config["api_key"], model_config["llm_name"], base_url=model_config["api_base"])
+
+        if llm_type == LLMType.RERANK:
+            if model_config["llm_factory"] not in RerankModel:
+                return
+            return RerankModel[model_config["llm_factory"]](model_config["api_key"], model_config["llm_name"], base_url=model_config["api_base"])
+
+        if llm_type == LLMType.IMAGE2TEXT.value:
+            if model_config["llm_factory"] not in CvModel:
+                return
+            return CvModel[model_config["llm_factory"]](model_config["api_key"], model_config["llm_name"], lang, base_url=model_config["api_base"], **kwargs)
+
+        if llm_type == LLMType.CHAT.value:
+            if model_config["llm_factory"] not in ChatModel:
+                return
+            return ChatModel[model_config["llm_factory"]](model_config["api_key"], model_config["llm_name"], base_url=model_config["api_base"], **kwargs)
+
+        if llm_type == LLMType.SPEECH2TEXT:
+            if model_config["llm_factory"] not in Seq2txtModel:
+                return
+            return Seq2txtModel[model_config["llm_factory"]](key=model_config["api_key"], model_name=model_config["llm_name"], lang=lang, base_url=model_config["api_base"])
+        if llm_type == LLMType.TTS:
+            if model_config["llm_factory"] not in TTSModel:
+                return
+            return TTSModel[model_config["llm_factory"]](
+                model_config["api_key"],
+                model_config["llm_name"],
+                base_url=model_config["api_base"],
+            )
+
+    @classmethod
+    @DB.connection_context()
+    def increase_usage(cls, tenant_id, llm_type, used_tokens, llm_name=None):
+        e, tenant = TenantService.get_by_id(tenant_id)
+        if not e:
+            logging.error(f"Tenant not found: {tenant_id}")
+            return 0
+
+        llm_map = {
+            LLMType.EMBEDDING.value: tenant.embd_id if not llm_name else llm_name,
+            LLMType.SPEECH2TEXT.value: tenant.asr_id,
+            LLMType.IMAGE2TEXT.value: tenant.img2txt_id,
+            LLMType.CHAT.value: tenant.llm_id if not llm_name else llm_name,
+            LLMType.RERANK.value: tenant.rerank_id if not llm_name else llm_name,
+            LLMType.TTS.value: tenant.tts_id if not llm_name else llm_name,
+        }
+
+        mdlnm = llm_map.get(llm_type)
+        if mdlnm is None:
+            logging.error(f"LLM type error: {llm_type}")
+            return 0
+
+        llm_name, llm_factory = TenantLLMService.split_model_name_and_factory(mdlnm)
+
+        try:
+            num = (
+                cls.model.update(used_tokens=cls.model.used_tokens + used_tokens)
+                .where(cls.model.tenant_id == tenant_id, cls.model.llm_name == llm_name, cls.model.llm_factory == llm_factory if llm_factory else True)
+                .execute()
+            )
+        except Exception:
+            logging.exception("TenantLLMService.increase_usage got exception,Failed to update used_tokens for tenant_id=%s, llm_name=%s", tenant_id, llm_name)
+            return 0
+
+        return num
+
+    @classmethod
+    @DB.connection_context()
+    def get_openai_models(cls):
+        objs = cls.model.select().where((cls.model.llm_factory == "OpenAI"), ~(cls.model.llm_name == "text-embedding-3-small"), ~(cls.model.llm_name == "text-embedding-3-large")).dicts()
+        return list(objs)
+
+    @staticmethod
+    def llm_id2llm_type(llm_id: str) -> str | None:
+        from api.db.services.llm_service import LLMService
+        llm_id, *_ = TenantLLMService.split_model_name_and_factory(llm_id)
+        llm_factories = settings.FACTORY_LLM_INFOS
+        for llm_factory in llm_factories:
+            for llm in llm_factory["llm"]:
+                if llm_id == llm["llm_name"]:
+                    return llm["model_type"].split(",")[-1]
+
+        for llm in LLMService.query(llm_name=llm_id):
+            return llm.model_type
+
+        llm = TenantLLMService.get_or_none(llm_name=llm_id)
+        if llm:
+            return llm.model_type
+        for llm in TenantLLMService.query(llm_name=llm_id):
+            return llm.model_type
+
+
+class LLM4Tenant:
+    def __init__(self, tenant_id, llm_type, llm_name=None, lang="Chinese", **kwargs):
+        self.tenant_id = tenant_id
+        self.llm_type = llm_type
+        self.llm_name = llm_name
+        self.mdl = TenantLLMService.model_instance(tenant_id, llm_type, llm_name, lang=lang, **kwargs)
+        assert self.mdl, "Can't find model for {}/{}/{}".format(tenant_id, llm_type, llm_name)
+        model_config = TenantLLMService.get_model_config(tenant_id, llm_type, llm_name)
+        self.max_length = model_config.get("max_tokens", 8192)
+
+        self.is_tools = model_config.get("is_tools", False)
+        self.verbose_tool_use = kwargs.get("verbose_tool_use")
+
+        langfuse_keys = TenantLangfuseService.filter_by_tenant(tenant_id=tenant_id)
+        self.langfuse = None
+        if langfuse_keys:
+            langfuse = Langfuse(public_key=langfuse_keys.public_key, secret_key=langfuse_keys.secret_key, host=langfuse_keys.host)
+            if langfuse.auth_check():
+                self.langfuse = langfuse
+                trace_id = self.langfuse.create_trace_id()
+                self.trace_context = {"trace_id": trace_id}

api/ragflow_server.py

@@ -33,7 +33,7 @@ import uuid
 
 from werkzeug.serving import run_simple
 from api import settings
-from api.apps import app
+from api.apps import app, smtp_mail_server
 from api.db.runtime_config import RuntimeConfig
 from api.db.services.document_service import DocumentService
 from api import utils
@@ -59,11 +59,14 @@ def update_progress():
             if redis_lock.acquire():
                 DocumentService.update_progress()
                 redis_lock.release()
-            stop_event.wait(6)
         except Exception:
             logging.exception("update_progress exception")
         finally:
-            redis_lock.release()
+            try:
+                redis_lock.release()
+            except Exception:
+                logging.exception("update_progress exception")
+            stop_event.wait(6)
 
 def signal_handler(sig, frame):
     logging.info("Received interrupt signal, shutting down...")
@@ -74,11 +77,11 @@ def signal_handler(sig, frame):
 
 if __name__ == '__main__':
     logging.info(r"""
-        ____   ___    ______ ______ __               
+        ____   ___    ______ ______ __
        / __ \ /   |  / ____// ____// /____  _      __
       / /_/ // /| | / / __ / /_   / // __ \| | /| / /
-     / _, _// ___ |/ /_/ // __/  / // /_/ /| |/ |/ / 
-    /_/ |_|/_/  |_|\____//_/    /_/ \____/ |__/|__/                             
+     / _, _// ___ |/ /_/ // __/  / // /_/ /| |/ |/ /
+    /_/ |_|/_/  |_|\____//_/    /_/ \____/ |__/|__/
 
     """)
     logging.info(
@@ -137,6 +140,18 @@ if __name__ == '__main__':
     else:
         threading.Timer(1.0, delayed_start_update_progress).start()
 
+    # init smtp server
+    if settings.SMTP_CONF:
+        app.config["MAIL_SERVER"] = settings.MAIL_SERVER
+        app.config["MAIL_PORT"] = settings.MAIL_PORT
+        app.config["MAIL_USE_SSL"] = settings.MAIL_USE_SSL
+        app.config["MAIL_USE_TLS"] = settings.MAIL_USE_TLS
+        app.config["MAIL_USERNAME"] = settings.MAIL_USERNAME
+        app.config["MAIL_PASSWORD"] = settings.MAIL_PASSWORD
+        app.config["MAIL_DEFAULT_SENDER"] = settings.MAIL_DEFAULT_SENDER
+        smtp_mail_server.init_app(app)
+
+
     # start http server
     try:
         logging.info("RAGFlow HTTP server start...")

api/settings.py

@@ -38,6 +38,11 @@ EMBEDDING_MDL = ""
 RERANK_MDL = ""
 ASR_MDL = ""
 IMAGE2TEXT_MDL = ""
+CHAT_CFG = ""
+EMBEDDING_CFG = ""
+RERANK_CFG = ""
+ASR_CFG = ""
+IMAGE2TEXT_CFG = ""
 API_KEY = None
 PARSERS = None
 HOST_IP = None
@@ -70,26 +75,36 @@ REGISTER_ENABLED = 1
 # sandbox-executor-manager
 SANDBOX_ENABLED = 0
 SANDBOX_HOST = None
+STRONG_TEST_COUNT = int(os.environ.get("STRONG_TEST_COUNT", "8"))
 
 BUILTIN_EMBEDDING_MODELS = ["BAAI/bge-large-zh-v1.5@BAAI", "maidalun1020/bce-embedding-base_v1@Youdao"]
 
+SMTP_CONF = None
+MAIL_SERVER = ""
+MAIL_PORT = 000
+MAIL_USE_SSL= True
+MAIL_USE_TLS = False
+MAIL_USERNAME = ""
+MAIL_PASSWORD = ""
+MAIL_DEFAULT_SENDER = ()
+MAIL_FRONTEND_URL = ""
+
+
 def get_or_create_secret_key():
     secret_key = os.environ.get("RAGFLOW_SECRET_KEY")
     if secret_key and len(secret_key) >= 32:
         return secret_key
-    
+
     # Check if there's a configured secret key
     configured_key = get_base_config(RAG_FLOW_SERVICE_NAME, {}).get("secret_key")
     if configured_key and configured_key != str(date.today()) and len(configured_key) >= 32:
         return configured_key
-    
+
     # Generate a new secure key and warn about it
     import logging
+
     new_key = secrets.token_hex(32)
-    logging.warning(
-        "SECURITY WARNING: Using auto-generated SECRET_KEY. "
-        f"Generated key: {new_key}"
-    )
+    logging.warning(f"SECURITY WARNING: Using auto-generated SECRET_KEY. Generated key: {new_key}")
     return new_key
 
 
@@ -98,10 +113,10 @@ def init_settings():
     LIGHTEN = int(os.environ.get("LIGHTEN", "0"))
     DATABASE_TYPE = os.getenv("DB_TYPE", "mysql")
     DATABASE = decrypt_database_config(name=DATABASE_TYPE)
-    LLM = get_base_config("user_default_llm", {})
-    LLM_DEFAULT_MODELS = LLM.get("default_models", {})
-    LLM_FACTORY = LLM.get("factory")
-    LLM_BASE_URL = LLM.get("base_url")
+    LLM = get_base_config("user_default_llm", {}) or {}
+    LLM_DEFAULT_MODELS = LLM.get("default_models", {}) or {}
+    LLM_FACTORY = LLM.get("factory", "") or ""
+    LLM_BASE_URL = LLM.get("base_url", "") or ""
     try:
         REGISTER_ENABLED = int(os.environ.get("REGISTER_ENABLED", "1"))
     except Exception:
@@ -114,29 +129,34 @@ def init_settings():
         FACTORY_LLM_INFOS = []
 
     global CHAT_MDL, EMBEDDING_MDL, RERANK_MDL, ASR_MDL, IMAGE2TEXT_MDL
+    global CHAT_CFG, EMBEDDING_CFG, RERANK_CFG, ASR_CFG, IMAGE2TEXT_CFG
     if not LIGHTEN:
         EMBEDDING_MDL = BUILTIN_EMBEDDING_MODELS[0]
 
-    if LLM_DEFAULT_MODELS:
-        CHAT_MDL = LLM_DEFAULT_MODELS.get("chat_model", CHAT_MDL)
-        EMBEDDING_MDL = LLM_DEFAULT_MODELS.get("embedding_model", EMBEDDING_MDL)
-        RERANK_MDL = LLM_DEFAULT_MODELS.get("rerank_model", RERANK_MDL)
-        ASR_MDL = LLM_DEFAULT_MODELS.get("asr_model", ASR_MDL)
-        IMAGE2TEXT_MDL = LLM_DEFAULT_MODELS.get("image2text_model", IMAGE2TEXT_MDL)
-
-        # factory can be specified in the config name with "@". LLM_FACTORY will be used if not specified
-        CHAT_MDL = CHAT_MDL + (f"@{LLM_FACTORY}" if "@" not in CHAT_MDL and CHAT_MDL != "" else "")
-        EMBEDDING_MDL = EMBEDDING_MDL + (f"@{LLM_FACTORY}" if "@" not in EMBEDDING_MDL and EMBEDDING_MDL != "" else "")
-        RERANK_MDL = RERANK_MDL + (f"@{LLM_FACTORY}" if "@" not in RERANK_MDL and RERANK_MDL != "" else "")
-        ASR_MDL = ASR_MDL + (f"@{LLM_FACTORY}" if "@" not in ASR_MDL and ASR_MDL != "" else "")
-        IMAGE2TEXT_MDL = IMAGE2TEXT_MDL + (f"@{LLM_FACTORY}" if "@" not in IMAGE2TEXT_MDL and IMAGE2TEXT_MDL != "" else "")
-
     global API_KEY, PARSERS, HOST_IP, HOST_PORT, SECRET_KEY
     API_KEY = LLM.get("api_key")
     PARSERS = LLM.get(
         "parsers", "naive:General,qa:Q&A,resume:Resume,manual:Manual,table:Table,paper:Paper,book:Book,laws:Laws,presentation:Presentation,picture:Picture,one:One,audio:Audio,email:Email,tag:Tag"
     )
 
+    chat_entry = _parse_model_entry(LLM_DEFAULT_MODELS.get("chat_model", CHAT_MDL))
+    embedding_entry = _parse_model_entry(LLM_DEFAULT_MODELS.get("embedding_model", EMBEDDING_MDL))
+    rerank_entry = _parse_model_entry(LLM_DEFAULT_MODELS.get("rerank_model", RERANK_MDL))
+    asr_entry = _parse_model_entry(LLM_DEFAULT_MODELS.get("asr_model", ASR_MDL))
+    image2text_entry = _parse_model_entry(LLM_DEFAULT_MODELS.get("image2text_model", IMAGE2TEXT_MDL))
+
+    CHAT_CFG = _resolve_per_model_config(chat_entry, LLM_FACTORY, API_KEY, LLM_BASE_URL)
+    EMBEDDING_CFG = _resolve_per_model_config(embedding_entry, LLM_FACTORY, API_KEY, LLM_BASE_URL)
+    RERANK_CFG = _resolve_per_model_config(rerank_entry, LLM_FACTORY, API_KEY, LLM_BASE_URL)
+    ASR_CFG = _resolve_per_model_config(asr_entry, LLM_FACTORY, API_KEY, LLM_BASE_URL)
+    IMAGE2TEXT_CFG = _resolve_per_model_config(image2text_entry, LLM_FACTORY, API_KEY, LLM_BASE_URL)
+
+    CHAT_MDL = CHAT_CFG.get("model", "") or ""
+    EMBEDDING_MDL = EMBEDDING_CFG.get("model", "") or ""
+    RERANK_MDL = RERANK_CFG.get("model", "") or ""
+    ASR_MDL = ASR_CFG.get("model", "") or ""
+    IMAGE2TEXT_MDL = IMAGE2TEXT_CFG.get("model", "") or ""
+
     HOST_IP = get_base_config(RAG_FLOW_SERVICE_NAME, {}).get("host", "127.0.0.1")
     HOST_PORT = get_base_config(RAG_FLOW_SERVICE_NAME, {}).get("http_port")
 
@@ -169,12 +189,28 @@ def init_settings():
 
     retrievaler = search.Dealer(docStoreConn)
     from graphrag import search as kg_search
+
     kg_retrievaler = kg_search.KGSearch(docStoreConn)
 
     if int(os.environ.get("SANDBOX_ENABLED", "0")):
         global SANDBOX_HOST
         SANDBOX_HOST = os.environ.get("SANDBOX_HOST", "sandbox-executor-manager")
 
+    global SMTP_CONF, MAIL_SERVER, MAIL_PORT, MAIL_USE_SSL, MAIL_USE_TLS
+    global MAIL_USERNAME, MAIL_PASSWORD, MAIL_DEFAULT_SENDER, MAIL_FRONTEND_URL
+    SMTP_CONF = get_base_config("smtp", {})
+
+    MAIL_SERVER = SMTP_CONF.get("mail_server", "")
+    MAIL_PORT = SMTP_CONF.get("mail_port", 000)
+    MAIL_USE_SSL = SMTP_CONF.get("mail_use_ssl", True)
+    MAIL_USE_TLS = SMTP_CONF.get("mail_use_tls", False)
+    MAIL_USERNAME = SMTP_CONF.get("mail_username", "")
+    MAIL_PASSWORD = SMTP_CONF.get("mail_password", "")
+    mail_default_sender = SMTP_CONF.get("mail_default_sender", [])
+    if mail_default_sender and len(mail_default_sender) >= 2:
+        MAIL_DEFAULT_SENDER = (mail_default_sender[0], mail_default_sender[1])
+    MAIL_FRONTEND_URL = SMTP_CONF.get("mail_frontend_url", "")
+
 
 class CustomEnum(Enum):
     @classmethod
@@ -209,3 +245,34 @@ class RetCode(IntEnum, CustomEnum):
     SERVER_ERROR = 500
     FORBIDDEN = 403
     NOT_FOUND = 404
+
+
+def _parse_model_entry(entry):
+    if isinstance(entry, str):
+        return {"name": entry, "factory": None, "api_key": None, "base_url": None}
+    if isinstance(entry, dict):
+        name = entry.get("name") or entry.get("model") or ""
+        return {
+            "name": name,
+            "factory": entry.get("factory"),
+            "api_key": entry.get("api_key"),
+            "base_url": entry.get("base_url"),
+        }
+    return {"name": "", "factory": None, "api_key": None, "base_url": None}
+
+
+def _resolve_per_model_config(entry_dict, backup_factory, backup_api_key, backup_base_url):
+    name = (entry_dict.get("name") or "").strip()
+    m_factory = entry_dict.get("factory") or backup_factory or ""
+    m_api_key = entry_dict.get("api_key") or backup_api_key or ""
+    m_base_url = entry_dict.get("base_url") or backup_base_url or ""
+
+    if name and "@" not in name and m_factory:
+        name = f"{name}@{m_factory}"
+
+    return {
+        "model": name,
+        "factory": m_factory,
+        "api_key": m_api_key,
+        "base_url": m_base_url,
+    }

api/utils/api_utils.py

@@ -17,6 +17,7 @@ import asyncio
 import functools
 import json
 import logging
+import os
 import queue
 import random
 import threading
@@ -48,7 +49,8 @@ from werkzeug.http import HTTP_STATUS_CODES
 from api import settings
 from api.constants import REQUEST_MAX_WAIT_SEC, REQUEST_WAIT_SEC
 from api.db.db_models import APIToken
-from api.db.services.llm_service import LLMService, TenantLLMService
+from api.db.services.llm_service import LLMService
+from api.db.services.tenant_llm_service import TenantLLMService
 from api.utils import CustomJSONEncoder, get_uuid, json_dumps
 from rag.utils.mcp_tool_call_conn import MCPToolCallSession, close_multiple_mcp_toolcall_sessions
 
@@ -352,7 +354,7 @@ def get_parser_config(chunk_method, parser_config):
     if not chunk_method:
         chunk_method = "naive"
 
-    # Define default configurations for each chunk method
+    # Define default configurations for each chunking method
     key_mapping = {
         "naive": {"chunk_token_num": 512, "delimiter": r"\n", "html4excel": False, "layout_recognize": "DeepDOC", "raptor": {"use_raptor": False}, "graphrag": {"use_graphrag": False}},
         "qa": {"raptor": {"use_raptor": False}, "graphrag": {"use_graphrag": False}},
@@ -402,8 +404,22 @@ def get_data_openai(
     finish_reason=None,
     object="chat.completion",
     param=None,
+    stream=False
 ):
     total_tokens = prompt_tokens + completion_tokens
+
+    if stream:
+        return {
+            "id": f"{id}",
+            "object": "chat.completion.chunk",
+            "model": model,
+            "choices": [{
+                "delta": {"content": content},
+                "finish_reason": finish_reason,
+                "index": 0,
+            }],
+        }
+
     return {
         "id": f"{id}",
         "object": object,
@@ -414,9 +430,21 @@ def get_data_openai(
             "prompt_tokens": prompt_tokens,
             "completion_tokens": completion_tokens,
             "total_tokens": total_tokens,
-            "completion_tokens_details": {"reasoning_tokens": 0, "accepted_prediction_tokens": 0, "rejected_prediction_tokens": 0},
+            "completion_tokens_details": {
+                "reasoning_tokens": 0,
+                "accepted_prediction_tokens": 0,
+                "rejected_prediction_tokens": 0,
+            },
         },
-        "choices": [{"message": {"role": "assistant", "content": content}, "logprobs": None, "finish_reason": finish_reason, "index": 0}],
+        "choices": [{
+            "message": {
+                "role": "assistant",
+                "content": content
+            },
+            "logprobs": None,
+            "finish_reason": finish_reason,
+            "index": 0,
+        }],
     }
 
 
@@ -640,7 +668,10 @@ def timeout(seconds: float | int = None, attempts: int = 2, *, exception: Option
 
             for a in range(attempts):
                 try:
-                    result = result_queue.get(timeout=seconds)
+                    if os.environ.get("ENABLE_TIMEOUT_ASSERTION"):
+                        result = result_queue.get(timeout=seconds)
+                    else:
+                        result = result_queue.get()
                     if isinstance(result, Exception):
                         raise result
                     return result
@@ -655,7 +686,10 @@ def timeout(seconds: float | int = None, attempts: int = 2, *, exception: Option
 
             for a in range(attempts):
                 try:
-                    with trio.fail_after(seconds):
+                    if os.environ.get("ENABLE_TIMEOUT_ASSERTION"):
+                        with trio.fail_after(seconds):
+                            return await func(*args, **kwargs)
+                    else:
                         return await func(*args, **kwargs)
                 except trio.TooSlowError:
                     if a < attempts - 1:
@@ -687,7 +721,13 @@ def timeout(seconds: float | int = None, attempts: int = 2, *, exception: Option
 
 
 async def is_strong_enough(chat_model, embedding_model):
-    @timeout(30, 2)
+    count = settings.STRONG_TEST_COUNT
+    if not chat_model or not embedding_model:
+        return
+    if isinstance(count, int) and count <= 0:
+        return
+
+    @timeout(60, 2)
     async def _is_strong_enough():
         nonlocal chat_model, embedding_model
         if embedding_model:
@@ -701,5 +741,5 @@ async def is_strong_enough(chat_model, embedding_model):
 
     # Pressure test for GraphRAG task
     async with trio.open_nursery() as nursery:
-        for _ in range(32):
+        for _ in range(count):
             nursery.start_soon(_is_strong_enough)

api/utils/base64_image.py

@@ -1 +1,3 @@
+import base64
 test_image_base64 = "iVBORw0KGgoAAAANSUhEUgAAAGQAAABkCAIAAAD/gAIDAAAA6ElEQVR4nO3QwQ3AIBDAsIP9d25XIC+EZE8QZc18w5l9O+AlZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBWYFZgVmBT+IYAHHLHkdEgAAAABJRU5ErkJggg=="
+test_image = base64.b64decode(test_image_base64)

api/utils/web_utils.py

@@ -21,6 +21,9 @@ import re
 import socket
 from urllib.parse import urlparse
 
+from api.apps import smtp_mail_server
+from flask_mail import Message
+from flask import render_template_string
 from selenium import webdriver
 from selenium.common.exceptions import TimeoutException
 from selenium.webdriver.chrome.options import Options
@@ -31,6 +34,7 @@ from selenium.webdriver.support.ui import WebDriverWait
 from webdriver_manager.chrome import ChromeDriverManager
 
 
+
 CONTENT_TYPE_MAP = {
     # Office
     "docx": "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
@@ -172,3 +176,26 @@ def get_float(req: dict, key: str, default: float | int = 10.0) -> float:
         return parsed if parsed > 0 else default
     except (TypeError, ValueError):
         return default
+
+
+INVITE_EMAIL_TMPL = """
+<p>Hi {{email}},</p>
+<p>{{inviter}} has invited you to join their team (ID: {{tenant_id}}).</p>
+<p>Click the link below to complete your registration:<br>
+<a href="{{invite_url}}">{{invite_url}}</a></p>
+<p>If you did not request this, please ignore this email.</p>
+"""
+
+def send_invite_email(to_email, invite_url, tenant_id, inviter):
+    from api.apps import  app
+    with app.app_context():
+        msg = Message(subject="RAGFlow Invitation",
+                      recipients=[to_email])
+        msg.html = render_template_string(
+            INVITE_EMAIL_TMPL,
+            email=to_email,
+            invite_url=invite_url,
+            tenant_id=tenant_id,
+            inviter=inviter,
+        )
+        smtp_mail_server.send(msg)

conf/llm_factories.json

@@ -6,6 +6,34 @@
             "tags": "LLM,TEXT EMBEDDING,TTS,TEXT RE-RANK,SPEECH2TEXT,MODERATION",
             "status": "1",
             "llm": [
+                {
+                    "llm_name": "gpt-5",
+                    "tags": "LLM,CHAT,400k,IMAGE2TEXT",
+                    "max_tokens": 400000,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "gpt-5-mini",
+                    "tags": "LLM,CHAT,400k,IMAGE2TEXT",
+                    "max_tokens": 400000,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "gpt-5-nano",
+                    "tags": "LLM,CHAT,400k,IMAGE2TEXT",
+                    "max_tokens": 400000,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "gpt-5-chat-latest",
+                    "tags": "LLM,CHAT,400k,IMAGE2TEXT",
+                    "max_tokens": 400000,
+                    "model_type": "chat",
+                    "is_tools": false
+                },
                 {
                     "llm_name": "gpt-4.1",
                     "tags": "LLM,CHAT,1M,IMAGE2TEXT",
@@ -477,6 +505,24 @@
                     "tags": "RE-RANK,4k",
                     "max_tokens": 4000,
                     "model_type": "rerank"
+                },
+                {
+                    "llm_name": "qwen-audio-asr",
+                    "tags": "SPEECH2TEXT,8k",
+                    "max_tokens": 8000,
+                    "model_type": "speech2text"
+                },
+                {
+                    "llm_name": "qwen-audio-asr-latest",
+                    "tags": "SPEECH2TEXT,8k",
+                    "max_tokens": 8000,
+                    "model_type": "speech2text"
+                },
+                {
+                    "llm_name": "qwen-audio-asr-1204",
+                    "tags": "SPEECH2TEXT,8k",
+                    "max_tokens": 8000,
+                    "model_type": "speech2text"
                 }
             ]
         },
@@ -486,23 +532,65 @@
             "tags": "LLM,TEXT EMBEDDING,SPEECH2TEXT,MODERATION",
             "status": "1",
             "llm": [
+                {
+                    "llm_name": "glm-4.5",
+                    "tags": "LLM,CHAT,128K",
+                    "max_tokens": 128000,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "glm-4.5-x",
+                    "tags": "LLM,CHAT,128k",
+                    "max_tokens": 128000,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "glm-4.5-air",
+                    "tags": "LLM,CHAT,128K",
+                    "max_tokens": 128000,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "glm-4.5-airx",
+                    "tags": "LLM,CHAT,128k",
+                    "max_tokens": 128000,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "glm-4.5-flash",
+                    "tags": "LLM,CHAT,128k",
+                    "max_tokens": 128000,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "glm-4.5v",
+                    "tags": "LLM,IMAGE2TEXT,64,",
+                    "max_tokens": 64000,
+                    "model_type": "image2text",
+                    "is_tools": false
+                },
                 {
                     "llm_name": "glm-4-plus",
-                    "tags": "LLM,CHAT,",
+                    "tags": "LLM,CHAT,128K",
                     "max_tokens": 128000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "glm-4-0520",
-                    "tags": "LLM,CHAT,",
+                    "tags": "LLM,CHAT,128K",
                     "max_tokens": 128000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "glm-4",
-                    "tags": "LLM,CHAT,",
+                    "tags":"LLM,CHAT,128K",
                     "max_tokens": 128000,
                     "model_type": "chat",
                     "is_tools": true
@@ -1118,60 +1206,35 @@
                     "llm_name": "gemini-2.5-flash",
                     "tags": "LLM,CHAT,1024K,IMAGE2TEXT",
                     "max_tokens": 1048576,
-                    "model_type": "image2text",
+                    "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "gemini-2.5-pro",
                     "tags": "LLM,CHAT,IMAGE2TEXT,1024K",
                     "max_tokens": 1048576,
-                    "model_type": "image2text",
+                    "model_type": "chat",
                     "is_tools": true
                 },
                 {
-                    "llm_name": "gemini-2.5-flash-preview-05-20",
+                    "llm_name": "gemini-2.5-flash-lite",
                     "tags": "LLM,CHAT,1024K,IMAGE2TEXT",
                     "max_tokens": 1048576,
-                    "model_type": "image2text",
+                    "model_type": "chat",
                     "is_tools": true
                 },
                 {
-                    "llm_name": "gemini-2.0-flash-001",
-                    "tags": "LLM,CHAT,1024K",
-                    "max_tokens": 1048576,
-                    "model_type": "image2text",
-                    "is_tools": true
-                },
-                {
-                    "llm_name": "gemini-2.0-flash-thinking-exp-01-21",
+                    "llm_name": "gemini-2.0-flash",
                     "tags": "LLM,CHAT,1024K",
                     "max_tokens": 1048576,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
-                    "llm_name": "gemini-1.5-flash",
-                    "tags": "LLM,IMAGE2TEXT,1024K",
+                    "llm_name": "gemini-2.0-flash-lite",
+                    "tags": "LLM,CHAT,1024K",
                     "max_tokens": 1048576,
-                    "model_type": "image2text"
-                },
-                {
-                    "llm_name": "gemini-2.5-pro-preview-05-06",
-                    "tags": "LLM,IMAGE2TEXT,1024K",
-                    "max_tokens": 1048576,
-                    "model_type": "image2text"
-                },
-                {
-                    "llm_name": "gemini-1.5-pro",
-                    "tags": "LLM,IMAGE2TEXT,2048K",
-                    "max_tokens": 2097152,
-                    "model_type": "image2text"
-                },
-                {
-                    "llm_name": "gemini-1.5-flash-8b",
-                    "tags": "LLM,IMAGE2TEXT,1024K",
-                    "max_tokens": 1048576,
-                    "model_type": "image2text",
+                    "model_type": "chat",
                     "is_tools": true
                 },
                 {
@@ -2598,234 +2661,255 @@
             "tags": "LLM,TEXT EMBEDDING,TEXT RE-RANK,IMAGE2TEXT",
             "status": "1",
             "llm": [
+                {
+                    "llm_name": "Qwen3-Embedding-8B",
+                    "tags": "TEXT EMBEDDING,TEXT RE-RANK,32k",
+                    "max_tokens": 32000,
+                    "model_type": "embedding",
+                    "is_tools": false
+                },
+                {
+                    "llm_name": "Qwen3-Embedding-4B",
+                    "tags": "TEXT EMBEDDING,TEXT RE-RANK,32k",
+                    "max_tokens": 32000,
+                    "model_type": "embedding",
+                    "is_tools": false
+                },
+                {
+                    "llm_name": "Qwen3-Embedding-0.6B",
+                    "tags": "TEXT EMBEDDING,TEXT RE-RANK,32k",
+                    "max_tokens": 32000,
+                    "model_type": "embedding",
+                    "is_tools": false
+                },
                 {
                     "llm_name": "Qwen/Qwen3-235B-A22B",
                     "tags": "LLM,CHAT,128k",
-                    "max_tokens": 8192,
+                    "max_tokens": 128000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/Qwen3-30B-A3B",
                     "tags": "LLM,CHAT,128k",
-                    "max_tokens": 8192,
+                    "max_tokens": 128000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/Qwen3-32B",
                     "tags": "LLM,CHAT,128k",
-                    "max_tokens": 8192,
+                    "max_tokens": 128000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/Qwen3-14B",
                     "tags": "LLM,CHAT,128k",
-                    "max_tokens": 8192,
+                    "max_tokens": 128000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/Qwen3-8B",
                     "tags": "LLM,CHAT,64k",
-                    "max_tokens": 8192,
+                    "max_tokens": 64000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/QVQ-72B-Preview",
                     "tags": "LLM,CHAT,IMAGE2TEXT,32k",
-                    "max_tokens": 16384,
+                    "max_tokens": 32000,
                     "model_type": "image2text",
                     "is_tools": false
                 },
                 {
                     "llm_name": "Pro/deepseek-ai/DeepSeek-R1",
                     "tags": "LLM,CHAT,64k",
-                    "max_tokens": 16384,
+                    "max_tokens": 64000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "deepseek-ai/DeepSeek-R1",
                     "tags": "LLM,CHAT,64k",
-                    "max_tokens": 16384,
+                    "max_tokens": 64000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Pro/deepseek-ai/DeepSeek-V3",
                     "tags": "LLM,CHAT,64k",
-                    "max_tokens": 8192,
+                    "max_tokens": 64000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "deepseek-ai/DeepSeek-V3",
                     "tags": "LLM,CHAT,64k",
-                    "max_tokens": 8192,
+                    "max_tokens": 64000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Pro/deepseek-ai/DeepSeek-V3-1226",
                     "tags": "LLM,CHAT,64k",
-                    "max_tokens": 4096,
+                    "max_tokens": 64000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 16384,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 16384,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Pro/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 16384,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 16384,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Pro/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 16384,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 16384,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "deepseek-ai/DeepSeek-V2.5",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/QwQ-32B",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 32768,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/Qwen2.5-VL-72B-Instruct",
                     "tags": "LLM,CHAT,IMAGE2TEXT,128k",
-                    "max_tokens": 4096,
+                    "max_tokens": 128000,
                     "model_type": "image2text",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Pro/Qwen/Qwen2.5-VL-7B-Instruct",
                     "tags": "LLM,CHAT,IMAGE2TEXT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "image2text",
                     "is_tools": true
                 },
                 {
                     "llm_name": "THUDM/GLM-Z1-32B-0414",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "THUDM/GLM-4-32B-0414",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 8192,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "THUDM/GLM-Z1-9B-0414",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 8192,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "THUDM/GLM-4-9B-0414",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "THUDM/chatglm3-6b",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": false
                 },
                 {
                     "llm_name": "Pro/THUDM/glm-4-9b-chat",
                     "tags": "LLM,CHAT,128k",
-                    "max_tokens": 4096,
+                    "max_tokens": 128000,
                     "model_type": "chat",
                     "is_tools": false
                 },
                 {
                     "llm_name": "THUDM/GLM-Z1-Rumination-32B-0414",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": false
                 },
                 {
                     "llm_name": "THUDM/glm-4-9b-chat",
                     "tags": "LLM,CHAT,128k",
-                    "max_tokens": 4096,
+                    "max_tokens": 128000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/QwQ-32B-Preview",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 8192,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": false
                 },
                 {
                     "llm_name": "Qwen/Qwen2.5-Coder-32B-Instruct",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": false
                 },
                 {
                     "llm_name": "Qwen/Qwen2-VL-72B-Instruct",
                     "tags": "LLM,IMAGE2TEXT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "image2text",
                     "is_tools": false
                 },
                 {
                     "llm_name": "Qwen/Qwen2.5-72B-Instruct-128Kt",
                     "tags": "LLM,IMAGE2TEXT,128k",
-                    "max_tokens": 4096,
+                    "max_tokens": 128000,
                     "model_type": "image2text",
                     "is_tools": false
                 },
@@ -2839,98 +2923,98 @@
                 {
                     "llm_name": "Qwen/Qwen2.5-72B-Instruct",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/Qwen2.5-32B-Instruct",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/Qwen2.5-14B-Instruct",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/Qwen2.5-7B-Instruct",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/Qwen2.5-Coder-7B-Instruct",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "internlm/internlm2_5-20b-chat",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "internlm/internlm2_5-7b-chat",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/Qwen2-7B-Instruct",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Qwen/Qwen2-1.5B-Instruct",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Pro/Qwen/Qwen2.5-Coder-7B-Instruct",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": false
                 },
                 {
                     "llm_name": "Pro/Qwen/Qwen2-VL-7B-Instruct",
                     "tags": "LLM,CHAT,IMAGE2TEXT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "image2text",
                     "is_tools": false
                 },
                 {
                     "llm_name": "Pro/Qwen/Qwen2.5-7B-Instruct",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "Pro/Qwen/Qwen2-7B-Instruct",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": false
                 },
                 {
                     "llm_name": "Pro/Qwen/Qwen2-1.5B-Instruct",
                     "tags": "LLM,CHAT,32k",
-                    "max_tokens": 4096,
+                    "max_tokens": 32000,
                     "model_type": "chat",
                     "is_tools": false
                 },
@@ -3267,45 +3351,52 @@
             "status": "1",
             "llm": [
                 {
-                    "llm_name": "claude-opus-4-20250514",
-                    "tags": "LLM,IMAGE2TEXT,200k",
-                    "max_tokens": 204800,
-                    "model_type": "image2text",
-                    "is_tools": true
-                },
-                {
-                    "llm_name": "claude-sonnet-4-20250514",
-                    "tags": "LLM,IMAGE2TEXT,200k",
-                    "max_tokens": 204800,
-                    "model_type": "image2text",
-                    "is_tools": true
-                },
-                {
-                    "llm_name": "claude-3-7-sonnet-20250219",
-                    "tags": "LLM,IMAGE2TEXT,200k",
-                    "max_tokens": 204800,
-                    "model_type": "image2text",
-                    "is_tools": true
-                },
-                {
-                    "llm_name": "claude-3-5-sonnet-20241022",
-                    "tags": "LLM,IMAGE2TEXT,200k",
+                    "llm_name": "claude-opus-4-1-20250805",
+                    "tags": "LLM,CHAT,IMAGE2TEXT,200k",
                     "max_tokens": 204800,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
-                    "llm_name": "claude-3-opus-20240229",
-                    "tags": "LLM,IMAGE2TEXT,200k",
+                    "llm_name": "claude-opus-4-20250514",
+                    "tags": "LLM,CHAT,IMAGE2TEXT,200k",
+                    "max_tokens": 204800,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "claude-sonnet-4-20250514",
+                    "tags": "LLM,CHAT,IMAGE2TEXT,200k",
+                    "max_tokens": 204800,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "claude-3-7-sonnet-20250219",
+                    "tags": "LLM,CHAT,IMAGE2TEXT,200k",
+                    "max_tokens": 204800,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "claude-3-5-sonnet-20241022",
+                    "tags": "LLM,CHAT,IMAGE2TEXT,200k",
+                    "max_tokens": 204800,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "claude-3-5-haiku-20241022",
+                    "tags": "LLM,CHAT,IMAGE2TEXT,200k",
                     "max_tokens": 204800,
                     "model_type": "chat",
                     "is_tools": true
                 },
                 {
                     "llm_name": "claude-3-haiku-20240307",
-                    "tags": "LLM,IMAGE2TEXT,200k",
+                    "tags": "LLM,CHAT,IMAGE2TEXT,200k",
                     "max_tokens": 204800,
-                    "model_type": "image2text",
+                    "model_type": "chat",
                     "is_tools": true
                 }
             ]

conf/service_conf.yaml

@@ -64,9 +64,21 @@ redis:
 #   config:
 #     oss_table: 'opendal_storage'
 # user_default_llm:
-#   factory: 'Tongyi-Qianwen'
-#   api_key: 'sk-xxxxxxxxxxxxx'
-#   base_url: ''
+#   factory: 'BAAI'
+#   api_key: 'backup'
+#   base_url: 'backup_base_url'
+#   default_models:
+#     chat_model:
+#       name: 'qwen2.5-7b-instruct'
+#       factory: 'xxxx'
+#       api_key: 'xxxx'
+#       base_url: 'https://api.xx.com'
+#     embedding_model:
+#       name: 'bge-m3'
+#     rerank_model: 'bge-reranker-v2'
+#     asr_model:
+#       model: 'whisper-large-v3' # alias of name
+#     image2text_model: ''
 # oauth:
 #   oauth2:
 #     display_name: "OAuth2"
@@ -101,3 +113,14 @@ redis:
 #   switch: false
 #   component: false
 #   dataset: false
+# smtp:
+#   mail_server: ""
+#   mail_port: 465
+#   mail_use_ssl: true
+#   mail_use_tls: false
+#   mail_username: ""
+#   mail_password: ""
+#   mail_default_sender:
+#     - "RAGFlow" # display name
+#     - "" # sender email address
+#   mail_frontend_url: "https://your-frontend.example.com"

deepdoc/parser/__init__.py

@@ -14,13 +14,15 @@
 #  limitations under the License.
 #
 
-from .pdf_parser import RAGFlowPdfParser as PdfParser, PlainParser
 from .docx_parser import RAGFlowDocxParser as DocxParser
 from .excel_parser import RAGFlowExcelParser as ExcelParser
-from .ppt_parser import RAGFlowPptParser as PptParser
 from .html_parser import RAGFlowHtmlParser as HtmlParser
 from .json_parser import RAGFlowJsonParser as JsonParser
+from .markdown_parser import MarkdownElementExtractor
 from .markdown_parser import RAGFlowMarkdownParser as MarkdownParser
+from .pdf_parser import PlainParser
+from .pdf_parser import RAGFlowPdfParser as PdfParser
+from .ppt_parser import RAGFlowPptParser as PptParser
 from .txt_parser import RAGFlowTxtParser as TxtParser
 
 __all__ = [
@@ -33,4 +35,6 @@ __all__ = [
     "JsonParser",
     "MarkdownParser",
     "TxtParser",
-]
+    "MarkdownElementExtractor",
+]
+

deepdoc/parser/excel_parser.py

@@ -12,6 +12,7 @@
 #
 
 import logging
+import re
 import sys
 from io import BytesIO
 
@@ -20,6 +21,8 @@ from openpyxl import Workbook, load_workbook
 
 from rag.nlp import find_codec
 
+# copied from `/openpyxl/cell/cell.py`
+ILLEGAL_CHARACTERS_RE = re.compile(r'[\000-\010]|[\013-\014]|[\016-\037]')
 
 class RAGFlowExcelParser:
 
@@ -50,13 +53,29 @@ class RAGFlowExcelParser:
             logging.info(f"openpyxl load error: {e}, try pandas instead")
             try:
                 file_like_object.seek(0)
-                df = pd.read_excel(file_like_object)
-                return RAGFlowExcelParser._dataframe_to_workbook(df)
+                try:
+                    df = pd.read_excel(file_like_object)
+                    return RAGFlowExcelParser._dataframe_to_workbook(df)
+                except Exception as ex:
+                    logging.info(f"pandas with default engine load error: {ex}, try calamine instead")
+                    file_like_object.seek(0)
+                    df = pd.read_excel(file_like_object, engine='calamine')
+                    return RAGFlowExcelParser._dataframe_to_workbook(df)
             except Exception as e_pandas:
                 raise Exception(f"pandas.read_excel error: {e_pandas}, original openpyxl error: {e}")
 
+    @staticmethod
+    def _clean_dataframe(df: pd.DataFrame):
+        def clean_string(s):
+            if isinstance(s, str):
+                return ILLEGAL_CHARACTERS_RE.sub(" ", s)
+            return s
+
+        return df.apply(lambda col: col.map(clean_string))
+
     @staticmethod
     def _dataframe_to_workbook(df):
+        df = RAGFlowExcelParser._clean_dataframe(df)
         wb = Workbook()
         ws = wb.active
         ws.title = "Data"
@@ -71,9 +90,17 @@ class RAGFlowExcelParser:
         return wb
 
     def html(self, fnm, chunk_rows=256):
+        from html import escape
+
         file_like_object = BytesIO(fnm) if not isinstance(fnm, str) else fnm
         wb = RAGFlowExcelParser._load_excel_to_workbook(file_like_object)
         tb_chunks = []
+
+        def _fmt(v):
+            if v is None:
+                return ""
+            return str(v).strip()
+
         for sheetname in wb.sheetnames:
             ws = wb[sheetname]
             rows = list(ws.rows)
@@ -82,7 +109,7 @@ class RAGFlowExcelParser:
 
             tb_rows_0 = "<tr>"
             for t in list(rows[0]):
-                tb_rows_0 += f"<th>{t.value}</th>"
+                tb_rows_0 += f"<th>{escape(_fmt(t.value))}</th>"
             tb_rows_0 += "</tr>"
 
             for chunk_i in range((len(rows) - 1) // chunk_rows + 1):
@@ -90,7 +117,7 @@ class RAGFlowExcelParser:
                 tb += f"<table><caption>{sheetname}</caption>"
                 tb += tb_rows_0
                 for r in list(
-                    rows[1 + chunk_i * chunk_rows: 1 + (chunk_i + 1) * chunk_rows]
+                    rows[1 + chunk_i * chunk_rows: min(1 + (chunk_i + 1) * chunk_rows, len(rows))]
                 ):
                     tb += "<tr>"
                     for i, c in enumerate(r):

deepdoc/parser/html_parser.py

@@ -15,35 +15,200 @@
 #  limitations under the License.
 #
 
-from rag.nlp import find_codec
-import readability
-import html_text
+from rag.nlp import find_codec, rag_tokenizer
+import uuid
 import chardet
-
+from bs4 import BeautifulSoup, NavigableString, Tag, Comment
+import html
 
 def get_encoding(file):
     with open(file,'rb') as f:
         tmp = chardet.detect(f.read())
         return tmp['encoding']
 
+BLOCK_TAGS = [
+    "h1", "h2", "h3", "h4", "h5", "h6",
+    "p", "div", "article", "section", "aside",
+    "ul", "ol", "li",
+    "table", "pre", "code", "blockquote",
+    "figure", "figcaption"
+]
+TITLE_TAGS = {"h1": "#", "h2": "##", "h3": "###", "h4": "#####", "h5": "#####", "h6": "######"}
+
 
 class RAGFlowHtmlParser:
-    def __call__(self, fnm, binary=None):
+    def __call__(self, fnm, binary=None, chunk_token_num=None):
         if binary:
             encoding = find_codec(binary)
             txt = binary.decode(encoding, errors="ignore")
         else:
             with open(fnm, "r",encoding=get_encoding(fnm)) as f:
                 txt = f.read()
-        return self.parser_txt(txt)
+        return self.parser_txt(txt, chunk_token_num)
 
     @classmethod
-    def parser_txt(cls, txt):
+    def parser_txt(cls, txt, chunk_token_num):
         if not isinstance(txt, str):
             raise TypeError("txt type should be string!")
-        html_doc = readability.Document(txt)
-        title = html_doc.title()
-        content = html_text.extract_text(html_doc.summary(html_partial=True))
-        txt = f"{title}\n{content}"
-        sections = txt.split("\n")
+
+        temp_sections = []
+        soup = BeautifulSoup(txt, "html5lib")
+        # delete <style> tag
+        for style_tag in soup.find_all(["style", "script"]):
+            style_tag.decompose()
+        # delete <script> tag in <div>
+        for div_tag in soup.find_all("div"):
+            for script_tag in div_tag.find_all("script"):
+                script_tag.decompose()
+        # delete inline style
+        for tag in soup.find_all(True):
+            if 'style' in tag.attrs:
+                del tag.attrs['style']
+        # delete HTML comment
+        for comment in soup.find_all(string=lambda text: isinstance(text, Comment)):
+            comment.extract()
+
+        cls.read_text_recursively(soup.body, temp_sections, chunk_token_num=chunk_token_num)
+        block_txt_list, table_list = cls.merge_block_text(temp_sections)
+        sections = cls.chunk_block(block_txt_list, chunk_token_num=chunk_token_num)
+        for table in table_list:
+            sections.append(table.get("content", ""))
         return sections
+
+    @classmethod
+    def split_table(cls, html_table, chunk_token_num=512):
+        soup = BeautifulSoup(html_table, "html.parser")
+        rows = soup.find_all("tr")
+        tables = []
+        current_table = []
+        current_count = 0
+        table_str_list = []
+        for row in rows:
+            tks_str = rag_tokenizer.tokenize(str(row))
+            token_count = len(tks_str.split(" ")) if tks_str else 0
+            if current_count + token_count > chunk_token_num:
+                tables.append(current_table)
+                current_table = []
+                current_count = 0
+            current_table.append(row)
+            current_count += token_count
+        if current_table:
+            tables.append(current_table)
+
+        for table_rows in tables:
+            new_table = soup.new_tag("table")
+            for row in table_rows:
+                new_table.append(row)
+            table_str_list.append(str(new_table))
+
+        return table_str_list
+
+    @classmethod
+    def read_text_recursively(cls, element, parser_result, chunk_token_num=512, parent_name=None, block_id=None):
+        if isinstance(element, NavigableString):
+            content = element.strip()
+
+            def is_valid_html(content):
+                try:
+                    soup = BeautifulSoup(content, "html.parser")
+                    return bool(soup.find())
+                except Exception:
+                    return False
+
+            return_info = []
+            if content:
+                if is_valid_html(content):
+                    soup = BeautifulSoup(content, "html.parser")
+                    child_info = cls.read_text_recursively(soup, parser_result, chunk_token_num, element.name, block_id)
+                    parser_result.extend(child_info)
+                else:
+                    info = {"content": element.strip(), "tag_name": "inner_text", "metadata": {"block_id": block_id}}
+                    if parent_name:
+                        info["tag_name"] = parent_name
+                    return_info.append(info)
+            return return_info
+        elif isinstance(element, Tag):
+
+            if str.lower(element.name) == "table":
+                table_info_list = []
+                table_id = str(uuid.uuid1())
+                table_list = [html.unescape(str(element))]
+                for t in table_list:
+                    table_info_list.append({"content": t, "tag_name": "table",
+                                            "metadata": {"table_id": table_id, "index": table_list.index(t)}})
+                return table_info_list
+            else:
+                block_id = None
+                if str.lower(element.name) in BLOCK_TAGS:
+                    block_id = str(uuid.uuid1())
+                for child in element.children:
+                    child_info = cls.read_text_recursively(child, parser_result, chunk_token_num, element.name,
+                                                           block_id)
+                    parser_result.extend(child_info)
+        return []
+
+    @classmethod
+    def merge_block_text(cls, parser_result):
+        block_content = []
+        current_content = ""
+        table_info_list = []
+        lask_block_id = None
+        for item in parser_result:
+            content = item.get("content")
+            tag_name = item.get("tag_name")
+            title_flag = tag_name in TITLE_TAGS
+            block_id = item.get("metadata", {}).get("block_id")
+            if block_id:
+                if title_flag:
+                    content = f"{TITLE_TAGS[tag_name]} {content}"
+                if lask_block_id != block_id:
+                    if lask_block_id is not None:
+                        block_content.append(current_content)
+                    current_content = content
+                    lask_block_id = block_id
+                else:
+                    current_content += (" " if current_content else "") + content
+            else:
+                if tag_name == "table":
+                    table_info_list.append(item)
+                else:
+                    current_content += (" " if current_content else "" + content)
+        if current_content:
+            block_content.append(current_content)
+        return block_content, table_info_list
+
+    @classmethod
+    def chunk_block(cls, block_txt_list, chunk_token_num=512):
+        chunks = []
+        current_block = ""
+        current_token_count = 0
+
+        for block in block_txt_list:
+            tks_str = rag_tokenizer.tokenize(block)
+            block_token_count = len(tks_str.split(" ")) if tks_str else 0
+            if block_token_count > chunk_token_num:
+                if current_block:
+                    chunks.append(current_block)
+                start = 0
+                tokens = tks_str.split(" ")
+                while start < len(tokens):
+                    end = start + chunk_token_num
+                    split_tokens = tokens[start:end]
+                    chunks.append(" ".join(split_tokens))
+                    start = end
+                current_block = ""
+                current_token_count = 0
+            else:
+                if current_token_count + block_token_count <= chunk_token_num:
+                    current_block += ("\n" if current_block else "") + block
+                    current_token_count += block_token_count
+                else:
+                    chunks.append(current_block)
+                    current_block = block
+                    current_token_count = block_token_count
+
+        if current_block:
+            chunks.append(current_block)
+
+        return chunks
+

deepdoc/parser/markdown_parser.py

@@ -17,8 +17,10 @@
 
 import re
 
+import mistune
 from markdown import markdown
 
+
 class RAGFlowMarkdownParser:
     def __init__(self, chunk_token_num=128):
         self.chunk_token_num = int(chunk_token_num)
@@ -35,40 +37,44 @@ class RAGFlowMarkdownParser:
                 table_list.append(raw_table)
                 if separate_tables:
                     # Skip this match (i.e., remove it)
-                    new_text += working_text[last_end:match.start()] + "\n\n"
+                    new_text += working_text[last_end : match.start()] + "\n\n"
                 else:
                     # Replace with rendered HTML
-                    html_table = markdown(raw_table, extensions=['markdown.extensions.tables']) if render else raw_table
-                    new_text += working_text[last_end:match.start()] + html_table + "\n\n"
+                    html_table = markdown(raw_table, extensions=["markdown.extensions.tables"]) if render else raw_table
+                    new_text += working_text[last_end : match.start()] + html_table + "\n\n"
                 last_end = match.end()
             new_text += working_text[last_end:]
             return new_text
 
-        if "|" in markdown_text: # for optimize performance
+        if "|" in markdown_text:  # for optimize performance
             # Standard Markdown table
             border_table_pattern = re.compile(
-                r'''
+                r"""
                 (?:\n|^)
                 (?:\|.*?\|.*?\|.*?\n)
                 (?:\|(?:\s*[:-]+[-| :]*\s*)\|.*?\n)
                 (?:\|.*?\|.*?\|.*?\n)+
-            ''', re.VERBOSE)
+            """,
+                re.VERBOSE,
+            )
             working_text = replace_tables_with_rendered_html(border_table_pattern, tables)
 
             # Borderless Markdown table
             no_border_table_pattern = re.compile(
-                r'''
+                r"""
                 (?:\n|^)
                 (?:\S.*?\|.*?\n)
                 (?:(?:\s*[:-]+[-| :]*\s*).*?\n)
                 (?:\S.*?\|.*?\n)+
-                ''', re.VERBOSE)
+                """,
+                re.VERBOSE,
+            )
             working_text = replace_tables_with_rendered_html(no_border_table_pattern, tables)
 
-        if "<table>" in working_text.lower(): # for optimize performance
-            #HTML table extraction - handle possible html/body wrapper tags
+        if "<table>" in working_text.lower():  # for optimize performance
+            # HTML table extraction - handle possible html/body wrapper tags
             html_table_pattern = re.compile(
-            r'''
+                r"""
             (?:\n|^)
             \s*
             (?:
@@ -83,9 +89,10 @@ class RAGFlowMarkdownParser:
             )
             \s*
             (?=\n|$)
-            ''',
-            re.VERBOSE | re.DOTALL | re.IGNORECASE
+            """,
+                re.VERBOSE | re.DOTALL | re.IGNORECASE,
             )
+
             def replace_html_tables():
                 nonlocal working_text
                 new_text = ""
@@ -94,9 +101,9 @@ class RAGFlowMarkdownParser:
                     raw_table = match.group()
                     tables.append(raw_table)
                     if separate_tables:
-                        new_text += working_text[last_end:match.start()] + "\n\n"
+                        new_text += working_text[last_end : match.start()] + "\n\n"
                     else:
-                        new_text += working_text[last_end:match.start()] + raw_table + "\n\n"
+                        new_text += working_text[last_end : match.start()] + raw_table + "\n\n"
                     last_end = match.end()
                 new_text += working_text[last_end:]
                 working_text = new_text
@@ -104,3 +111,163 @@ class RAGFlowMarkdownParser:
             replace_html_tables()
 
         return working_text, tables
+
+
+class MarkdownElementExtractor:
+    def __init__(self, markdown_content):
+        self.markdown_content = markdown_content
+        self.lines = markdown_content.split("\n")
+        self.ast_parser = mistune.create_markdown(renderer="ast")
+        self.ast_nodes = self.ast_parser(markdown_content)
+
+    def extract_elements(self):
+        """Extract individual elements (headers, code blocks, lists, etc.)"""
+        sections = []
+
+        i = 0
+        while i < len(self.lines):
+            line = self.lines[i]
+
+            if re.match(r"^#{1,6}\s+.*$", line):
+                # header
+                element = self._extract_header(i)
+                sections.append(element["content"])
+                i = element["end_line"] + 1
+            elif line.strip().startswith("```"):
+                # code block
+                element = self._extract_code_block(i)
+                sections.append(element["content"])
+                i = element["end_line"] + 1
+            elif re.match(r"^\s*[-*+]\s+.*$", line) or re.match(r"^\s*\d+\.\s+.*$", line):
+                # list block
+                element = self._extract_list_block(i)
+                sections.append(element["content"])
+                i = element["end_line"] + 1
+            elif line.strip().startswith(">"):
+                # blockquote
+                element = self._extract_blockquote(i)
+                sections.append(element["content"])
+                i = element["end_line"] + 1
+            elif line.strip():
+                # text block (paragraphs and inline elements until next block element)
+                element = self._extract_text_block(i)
+                sections.append(element["content"])
+                i = element["end_line"] + 1
+            else:
+                i += 1
+
+        sections = [section for section in sections if section.strip()]
+        return sections
+
+    def _extract_header(self, start_pos):
+        return {
+            "type": "header",
+            "content": self.lines[start_pos],
+            "start_line": start_pos,
+            "end_line": start_pos,
+        }
+
+    def _extract_code_block(self, start_pos):
+        end_pos = start_pos
+        content_lines = [self.lines[start_pos]]
+
+        # Find the end of the code block
+        for i in range(start_pos + 1, len(self.lines)):
+            content_lines.append(self.lines[i])
+            end_pos = i
+            if self.lines[i].strip().startswith("```"):
+                break
+
+        return {
+            "type": "code_block",
+            "content": "\n".join(content_lines),
+            "start_line": start_pos,
+            "end_line": end_pos,
+        }
+
+    def _extract_list_block(self, start_pos):
+        end_pos = start_pos
+        content_lines = []
+
+        i = start_pos
+        while i < len(self.lines):
+            line = self.lines[i]
+            # check if this line is a list item or continuation of a list
+            if (
+                re.match(r"^\s*[-*+]\s+.*$", line)
+                or re.match(r"^\s*\d+\.\s+.*$", line)
+                or (i > start_pos and not line.strip())
+                or (i > start_pos and re.match(r"^\s{2,}[-*+]\s+.*$", line))
+                or (i > start_pos and re.match(r"^\s{2,}\d+\.\s+.*$", line))
+                or (i > start_pos and re.match(r"^\s+\w+.*$", line))
+            ):
+                content_lines.append(line)
+                end_pos = i
+                i += 1
+            else:
+                break
+
+        return {
+            "type": "list_block",
+            "content": "\n".join(content_lines),
+            "start_line": start_pos,
+            "end_line": end_pos,
+        }
+
+    def _extract_blockquote(self, start_pos):
+        end_pos = start_pos
+        content_lines = []
+
+        i = start_pos
+        while i < len(self.lines):
+            line = self.lines[i]
+            if line.strip().startswith(">") or (i > start_pos and not line.strip()):
+                content_lines.append(line)
+                end_pos = i
+                i += 1
+            else:
+                break
+
+        return {
+            "type": "blockquote",
+            "content": "\n".join(content_lines),
+            "start_line": start_pos,
+            "end_line": end_pos,
+        }
+
+    def _extract_text_block(self, start_pos):
+        """Extract a text block (paragraphs, inline elements) until next block element"""
+        end_pos = start_pos
+        content_lines = [self.lines[start_pos]]
+
+        i = start_pos + 1
+        while i < len(self.lines):
+            line = self.lines[i]
+            # stop if we encounter a block element
+            if re.match(r"^#{1,6}\s+.*$", line) or line.strip().startswith("```") or re.match(r"^\s*[-*+]\s+.*$", line) or re.match(r"^\s*\d+\.\s+.*$", line) or line.strip().startswith(">"):
+                break
+            elif not line.strip():
+                # check if the next line is a block element
+                if i + 1 < len(self.lines) and (
+                    re.match(r"^#{1,6}\s+.*$", self.lines[i + 1])
+                    or self.lines[i + 1].strip().startswith("```")
+                    or re.match(r"^\s*[-*+]\s+.*$", self.lines[i + 1])
+                    or re.match(r"^\s*\d+\.\s+.*$", self.lines[i + 1])
+                    or self.lines[i + 1].strip().startswith(">")
+                ):
+                    break
+                else:
+                    content_lines.append(line)
+                    end_pos = i
+                    i += 1
+            else:
+                content_lines.append(line)
+                end_pos = i
+                i += 1
+
+        return {
+            "type": "text_block",
+            "content": "\n".join(content_lines),
+            "start_line": start_pos,
+            "end_line": end_pos,
+        }

deepdoc/parser/ppt_parser.py

@@ -87,7 +87,7 @@ class RAGFlowPptParser:
                 break
             texts = []
             for shape in sorted(
-                    slide.shapes, key=lambda x: ((x.top if x.top is not None else 0) // 10, x.left)):
+                    slide.shapes, key=lambda x: ((x.top if x.top is not None else 0) // 10, x.left if x.left is not None else 0)):
                 try:
                     txt = self.__extract(shape)
                     if txt:
@@ -96,4 +96,4 @@ class RAGFlowPptParser:
                     logging.exception(e)
             txts.append("\n".join(texts))
 
-        return txts
+        return txts

docker/.env

@@ -62,6 +62,8 @@ MYSQL_DBNAME=rag_flow
 # The port used to expose the MySQL service to the host machine,
 # allowing EXTERNAL access to the MySQL database running inside the Docker container.
 MYSQL_PORT=5455
+# The maximum size of communication packets sent to the MySQL server
+MYSQL_MAX_PACKET=1073741824
 
 # The hostname where the MinIO service is exposed
 MINIO_HOST=minio
@@ -91,13 +93,13 @@ REDIS_PASSWORD=infini_rag_flow
 SVR_HTTP_PORT=9380
 
 # The RAGFlow Docker image to download.
-# Defaults to the v0.20.0-slim edition, which is the RAGFlow Docker image without embedding models.
-RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.0-slim
+# Defaults to the v0.20.4-slim edition, which is the RAGFlow Docker image without embedding models.
+RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.4-slim
 #
 # To download the RAGFlow Docker image with embedding models, uncomment the following line instead:
-# RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.0
+# RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.4
 #
-# The Docker image of the v0.20.0 edition includes built-in embedding models:
+# The Docker image of the v0.20.4 edition includes built-in embedding models:
 #   - BAAI/bge-large-zh-v1.5
 #   - maidalun1020/bce-embedding-base_v1
 #

docker/README.md

@@ -79,8 +79,8 @@ The [.env](./.env) file contains important environment variables for Docker.
 - `RAGFLOW-IMAGE`  
   The Docker image edition. Available editions:  
   
-  - `infiniflow/ragflow:v0.20.0-slim` (default): The RAGFlow Docker image without embedding models.  
-  - `infiniflow/ragflow:v0.20.0`: The RAGFlow Docker image with embedding models including:
+  - `infiniflow/ragflow:v0.20.4-slim` (default): The RAGFlow Docker image without embedding models.  
+  - `infiniflow/ragflow:v0.20.4`: The RAGFlow Docker image with embedding models including:
     - Built-in embedding models:
       - `BAAI/bge-large-zh-v1.5` 
       - `maidalun1020/bce-embedding-base_v1`

docker/migration.sh

@@ -0,0 +1,298 @@
+#!/bin/bash
+
+# RAGFlow Data Migration Script
+# Usage: ./migration.sh [backup|restore] [backup_folder]
+# 
+# This script helps you backup and restore RAGFlow Docker volumes
+# including MySQL, MinIO, Redis, and Elasticsearch data.
+
+set -e  # Exit on any error
+# Instead, we'll handle errors manually for better debugging experience
+
+# Default values
+DEFAULT_BACKUP_FOLDER="backup"
+VOLUMES=("docker_mysql_data" "docker_minio_data" "docker_redis_data" "docker_esdata01")
+BACKUP_FILES=("mysql_backup.tar.gz" "minio_backup.tar.gz" "redis_backup.tar.gz" "es_backup.tar.gz")
+
+# Function to display help information
+show_help() {
+    echo "RAGFlow Data Migration Tool"
+    echo ""
+    echo "USAGE:"
+    echo "  $0 <operation> [backup_folder]"
+    echo ""
+    echo "OPERATIONS:"
+    echo "  backup   - Create backup of all RAGFlow data volumes"
+    echo "  restore  - Restore RAGFlow data volumes from backup"
+    echo "  help     - Show this help message"
+    echo ""
+    echo "PARAMETERS:"
+    echo "  backup_folder  - Name of backup folder (default: '$DEFAULT_BACKUP_FOLDER')"
+    echo ""
+    echo "EXAMPLES:"
+    echo "  $0 backup                    # Backup to './backup' folder"
+    echo "  $0 backup my_backup          # Backup to './my_backup' folder"
+    echo "  $0 restore                   # Restore from './backup' folder"
+    echo "  $0 restore my_backup         # Restore from './my_backup' folder"
+    echo ""
+    echo "DOCKER VOLUMES:"
+    echo "  - docker_mysql_data     (MySQL database)"
+    echo "  - docker_minio_data     (MinIO object storage)"
+    echo "  - docker_redis_data     (Redis cache)"
+    echo "  - docker_esdata01       (Elasticsearch indices)"
+}
+
+# Function to check if Docker is running
+check_docker() {
+    if ! docker info >/dev/null 2>&1; then
+        echo "❌ Error: Docker is not running or not accessible"
+        echo "Please start Docker and try again"
+        exit 1
+    fi
+}
+
+# Function to check if volume exists
+volume_exists() {
+    local volume_name=$1
+    docker volume inspect "$volume_name" >/dev/null 2>&1
+}
+
+# Function to check if any containers are using the target volumes
+check_containers_using_volumes() {
+    echo "🔍 Checking for running containers that might be using target volumes..."
+    
+    # Get all running containers
+    local running_containers=$(docker ps --format "{{.Names}}")
+    
+    if [ -z "$running_containers" ]; then
+        echo "✅ No running containers found"
+        return 0
+    fi
+    
+    # Check each running container for volume usage
+    local containers_using_volumes=()
+    local volume_usage_details=()
+    
+    for container in $running_containers; do
+        # Get container's mount information
+        local mounts=$(docker inspect "$container" --format '{{range .Mounts}}{{.Source}}{{"|"}}{{end}}' 2>/dev/null || echo "")
+        
+        # Check if any of our target volumes are used by this container
+        for volume in "${VOLUMES[@]}"; do
+            if echo "$mounts" | grep -q "$volume"; then
+                containers_using_volumes+=("$container")
+                volume_usage_details+=("$container -> $volume")
+                break
+            fi
+        done
+    done
+    
+    # If any containers are using our volumes, show error and exit
+    if [ ${#containers_using_volumes[@]} -gt 0 ]; then
+        echo ""
+        echo "❌ ERROR: Found running containers using target volumes!"
+        echo ""
+        echo "📋 Running containers status:"
+        docker ps --format "table {{.Names}}\t{{.Status}}\t{{.Image}}"
+        echo ""
+        echo "🔗 Volume usage details:"
+        for detail in "${volume_usage_details[@]}"; do
+            echo "  - $detail"
+        done
+        echo ""
+        echo "🛑 SOLUTION: Stop the containers before performing backup/restore operations:"
+        echo "   docker-compose -f docker/<your-docker-compose-file>.yml down"
+        echo ""
+        echo "💡 After backup/restore, you can restart with:"
+        echo "   docker-compose -f docker/<your-docker-compose-file>.yml up -d"
+        echo ""
+        exit 1
+    fi
+    
+    echo "✅ No containers are using target volumes, safe to proceed"
+    return 0
+}
+
+# Function to confirm user action
+confirm_action() {
+    local message=$1
+    echo -n "$message (y/N): "
+    read -r response
+    case "$response" in
+        [yY]|[yY][eE][sS]) return 0 ;;
+        *) return 1 ;;
+    esac
+}
+
+# Function to perform backup
+perform_backup() {
+    local backup_folder=$1
+    
+    echo "🚀 Starting RAGFlow data backup..."
+    echo "📁 Backup folder: $backup_folder"
+    echo ""
+    
+    # Check if any containers are using the volumes
+    check_containers_using_volumes
+    
+    # Create backup folder if it doesn't exist
+    mkdir -p "$backup_folder"
+    
+    # Backup each volume
+    for i in "${!VOLUMES[@]}"; do
+        local volume="${VOLUMES[$i]}"
+        local backup_file="${BACKUP_FILES[$i]}"
+        local step=$((i + 1))
+        
+        echo "📦 Step $step/4: Backing up $volume..."
+        
+        if volume_exists "$volume"; then
+            docker run --rm \
+                -v "$volume":/source \
+                -v "$(pwd)/$backup_folder":/backup \
+                alpine tar czf "/backup/$backup_file" -C /source .
+            echo "✅ Successfully backed up $volume to $backup_folder/$backup_file"
+        else
+            echo "⚠️  Warning: Volume $volume does not exist, skipping..."
+        fi
+        echo ""
+    done
+    
+    echo "🎉 Backup completed successfully!"
+    echo "📍 Backup location: $(pwd)/$backup_folder"
+    
+    # List backup files with sizes
+    echo ""
+    echo "📋 Backup files created:"
+    for backup_file in "${BACKUP_FILES[@]}"; do
+        if [ -f "$backup_folder/$backup_file" ]; then
+            local size=$(ls -lh "$backup_folder/$backup_file" | awk '{print $5}')
+            echo "  - $backup_file ($size)"
+        fi
+    done
+}
+
+# Function to perform restore
+perform_restore() {
+    local backup_folder=$1
+    
+    echo "🔄 Starting RAGFlow data restore..."
+    echo "📁 Backup folder: $backup_folder"
+    echo ""
+    
+    # Check if any containers are using the volumes
+    check_containers_using_volumes
+    
+    # Check if backup folder exists
+    if [ ! -d "$backup_folder" ]; then
+        echo "❌ Error: Backup folder '$backup_folder' does not exist"
+        exit 1
+    fi
+    
+    # Check if all backup files exist
+    local missing_files=()
+    for backup_file in "${BACKUP_FILES[@]}"; do
+        if [ ! -f "$backup_folder/$backup_file" ]; then
+            missing_files+=("$backup_file")
+        fi
+    done
+    
+    if [ ${#missing_files[@]} -gt 0 ]; then
+        echo "❌ Error: Missing backup files:"
+        for file in "${missing_files[@]}"; do
+            echo "  - $file"
+        done
+        echo "Please ensure all backup files are present in '$backup_folder'"
+        exit 1
+    fi
+    
+    # Check for existing volumes and warn user
+    local existing_volumes=()
+    for volume in "${VOLUMES[@]}"; do
+        if volume_exists "$volume"; then
+            existing_volumes+=("$volume")
+        fi
+    done
+    
+    if [ ${#existing_volumes[@]} -gt 0 ]; then
+        echo "⚠️  WARNING: The following Docker volumes already exist:"
+        for volume in "${existing_volumes[@]}"; do
+            echo "  - $volume"
+        done
+        echo ""
+        echo "🔴 IMPORTANT: Restoring will OVERWRITE existing data!"
+        echo "💡 Recommendation: Create a backup of your current data first:"
+        echo "   $0 backup current_backup_$(date +%Y%m%d_%H%M%S)"
+        echo ""
+        
+        if ! confirm_action "Do you want to continue with the restore operation?"; then
+            echo "❌ Restore operation cancelled by user"
+            exit 0
+        fi
+    fi
+    
+    # Create volumes and restore data
+    for i in "${!VOLUMES[@]}"; do
+        local volume="${VOLUMES[$i]}"
+        local backup_file="${BACKUP_FILES[$i]}"
+        local step=$((i + 1))
+        
+        echo "🔧 Step $step/4: Restoring $volume..."
+        
+        # Create volume if it doesn't exist
+        if ! volume_exists "$volume"; then
+            echo "  📋 Creating Docker volume: $volume"
+            docker volume create "$volume"
+        else
+            echo "  📋 Using existing Docker volume: $volume"
+        fi
+        
+        # Restore data
+        echo "  📥 Restoring data from $backup_file..."
+        docker run --rm \
+            -v "$volume":/target \
+            -v "$(pwd)/$backup_folder":/backup \
+            alpine tar xzf "/backup/$backup_file" -C /target
+        
+        echo "✅ Successfully restored $volume"
+        echo ""
+    done
+    
+    echo "🎉 Restore completed successfully!"
+    echo "💡 You can now start your RAGFlow services"
+}
+
+# Main script logic
+main() {
+    # Check if Docker is available
+    check_docker
+    
+    # Parse command line arguments
+    local operation=${1:-}
+    local backup_folder=${2:-$DEFAULT_BACKUP_FOLDER}
+    
+    # Handle help or no arguments
+    if [ -z "$operation" ] || [ "$operation" = "help" ] || [ "$operation" = "-h" ] || [ "$operation" = "--help" ]; then
+        show_help
+        exit 0
+    fi
+    
+    # Validate operation
+    case "$operation" in
+        backup)
+            perform_backup "$backup_folder"
+            ;;
+        restore)
+            perform_restore "$backup_folder"
+            ;;
+        *)
+            echo "❌ Error: Invalid operation '$operation'"
+            echo ""
+            show_help
+            exit 1
+            ;;
+    esac
+}
+
+# Run main function with all arguments
+main "$@"

docker/nginx/proxy.conf

@@ -6,3 +6,7 @@ proxy_set_header Connection "";
 proxy_buffering off;
 proxy_read_timeout 3600s;
 proxy_send_timeout 3600s;
+proxy_buffer_size 1024k;
+proxy_buffers 16 1024k;
+proxy_busy_buffers_size 2048k;
+proxy_temp_file_write_size 2048k;

docker/service_conf.yaml.template

@@ -9,6 +9,7 @@ mysql:
   port: 3306
   max_connections: 900
   stale_timeout: 300
+  max_allowed_packet: ${MYSQL_MAX_PACKET:-1073741824}
 minio:
   user: '${MINIO_USER:-rag_flow}'
   password: '${MINIO_PASSWORD:-infini_rag_flow}'

docs/configurations.md

@@ -99,8 +99,8 @@ RAGFlow utilizes MinIO as its object storage solution, leveraging its scalabilit
 - `RAGFLOW-IMAGE`  
   The Docker image edition. Available editions:  
   
-  - `infiniflow/ragflow:v0.20.0-slim` (default): The RAGFlow Docker image without embedding models.  
-  - `infiniflow/ragflow:v0.20.0`: The RAGFlow Docker image with embedding models including:
+  - `infiniflow/ragflow:v0.20.4-slim` (default): The RAGFlow Docker image without embedding models.  
+  - `infiniflow/ragflow:v0.20.4`: The RAGFlow Docker image with embedding models including:
     - Built-in embedding models:
       - `BAAI/bge-large-zh-v1.5` 
       - `maidalun1020/bce-embedding-base_v1`

docs/develop/acquire_ragflow_api_key.md

@@ -11,7 +11,7 @@ An API key is required for the RAGFlow server to authenticate your HTTP/Python o
 2. Click **API** to switch to the **API** page.
 3. Obtain a RAGFlow API key:
 
-![ragflow_api_key](https://github.com/user-attachments/assets/f461ed61-04c6-4faf-b3d8-6b5fa56be4e7)
+![ragflow_api_key](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/ragflow_api_key.jpg)
 
 :::tip NOTE
 See the [RAGFlow HTTP API reference](../references/http_api_reference.md) or the [RAGFlow Python API reference](../references/python_api_reference.md) for a complete reference of RAGFlow's HTTP or Python APIs.

docs/develop/build_docker_image.mdx

@@ -77,7 +77,7 @@ After building the infiniflow/ragflow:nightly-slim image, you are ready to launc
 
 1. Edit Docker Compose Configuration
 
-Open the `docker/.env` file. Find the `RAGFLOW_IMAGE` setting and change the image reference from `infiniflow/ragflow:v0.20.0-slim` to `infiniflow/ragflow:nightly-slim` to use the pre-built image.
+Open the `docker/.env` file. Find the `RAGFLOW_IMAGE` setting and change the image reference from `infiniflow/ragflow:v0.20.4-slim` to `infiniflow/ragflow:nightly-slim` to use the pre-built image.
 
 
 2. Launch the Service

docs/faq.mdx

@@ -30,17 +30,17 @@ The "garbage in garbage out" status quo remains unchanged despite the fact that
 
 Each RAGFlow release is available in two editions:
 
-- **Slim edition**: excludes built-in embedding models and is identified by a **-slim** suffix added to the version name. Example: `infiniflow/ragflow:v0.20.0-slim`
-- **Full edition**: includes built-in embedding models and has no suffix added to the version name. Example: `infiniflow/ragflow:v0.20.0`
+- **Slim edition**: excludes built-in embedding models and is identified by a **-slim** suffix added to the version name. Example: `infiniflow/ragflow:v0.20.4-slim`
+- **Full edition**: includes built-in embedding models and has no suffix added to the version name. Example: `infiniflow/ragflow:v0.20.4`
 
 ---
 
 ### Which embedding models can be deployed locally?
 
-RAGFlow offers two Docker image editions, `v0.20.0-slim` and `v0.20.0`:  
+RAGFlow offers two Docker image editions, `v0.20.4-slim` and `v0.20.4`:  
   
-- `infiniflow/ragflow:v0.20.0-slim` (default): The RAGFlow Docker image without embedding models.  
-- `infiniflow/ragflow:v0.20.0`: The RAGFlow Docker image with embedding models including:
+- `infiniflow/ragflow:v0.20.4-slim` (default): The RAGFlow Docker image without embedding models.  
+- `infiniflow/ragflow:v0.20.4`: The RAGFlow Docker image with embedding models including:
   - Built-in embedding models:
     - `BAAI/bge-large-zh-v1.5`
     - `maidalun1020/bce-embedding-base_v1`

docs/guides/agent/agent_component_reference/agent.mdx

@@ -9,7 +9,7 @@ The component equipped with reasoning, tool usage, and multi-agent collaboration
 
 ---
 
-An **Agent** component fine-tunes the LLM and sets its prompt. From v0.20.0 onwards, an **Agent** component is able to work independently and with the following capabilities:
+An **Agent** component fine-tunes the LLM and sets its prompt. From v0.20.4 onwards, an **Agent** component is able to work independently and with the following capabilities:
 
 - Autonomous reasoning with reflection and adjustment based on environmental feedback.
 - Use of tools or subagents to complete tasks.
@@ -82,7 +82,7 @@ An integer specifying the number of previous dialogue rounds to input into the L
 This feature is used for multi-turn dialogue *only*.
 :::
 
-### Max retrieves
+### Max retries
 
 Defines the maximum number of attempts the agent will make to retry a failed task or operation before stopping or reporting failure.
 
@@ -92,7 +92,11 @@ The waiting period in seconds that the agent observes before retrying a failed t
 
 ### Max rounds
 
-Defines the maximum number reflection rounds of the selected chat model. Defaults to 5 rounds.
+Defines the maximum number reflection rounds of the selected chat model. Defaults to 1 round.
+
+:::tip NOTE
+Increasing this value will significantly extend your agent's response time.
+:::
 
 ### Output
 

docs/guides/agent/agent_component_reference/retrieval.mdx

@@ -9,7 +9,7 @@ A component that retrieves information from specified datasets.
 
 ## Scenarios
 
-A **Retrieval** component is essential in most RAG scenarios, where information is extracted from designated knowledge bases before being sent to the LLM for content generation. As of v0.20.0, a **Retrieval** component can operate either as a workflow component or as a tool of an **Agent**, enabling the Agent to control its invocation and search queries.
+A **Retrieval** component is essential in most RAG scenarios, where information is extracted from designated knowledge bases before being sent to the LLM for content generation. As of v0.20.4, a **Retrieval** component can operate either as a workflow component or as a tool of an **Agent**, enabling the Agent to control its invocation and search queries.
 
 ## Configurations
 

docs/guides/agent/sandbox_quickstart.md

@@ -63,7 +63,7 @@ docker build -t sandbox-executor-manager:latest ./executor_manager
 3. Add the following entry to your /etc/hosts file to resolve the executor manager service:
 
 ```bash
-127.0.0.1 sandbox-executor-manager
+127.0.0.1 es01 infinity mysql minio redis sandbox-executor-manager
 ```
 
 4. Start the RAGFlow service as usual.

docs/guides/chat/start_chat.md

@@ -48,7 +48,7 @@ You start an AI conversation by creating an assistant.
      - If no target language is selected, the system will search only in the language of your query, which may cause relevant information in other languages to be missed.
    - **Variable** refers to the variables (keys) to be used in the system prompt. `{knowledge}` is a reserved variable. Click **Add** to add more variables for the system prompt.
       - If you are uncertain about the logic behind **Variable**, leave it *as-is*.
-      - As of v0.20.0, if you add custom variables here, the only way you can pass in their values is to call:
+      - As of v0.20.4, if you add custom variables here, the only way you can pass in their values is to call:
          - HTTP method [Converse with chat assistant](../../references/http_api_reference.md#converse-with-chat-assistant), or
          - Python method [Converse with chat assistant](../../references/python_api_reference.md#converse-with-chat-assistant).
 

docs/guides/dataset/configure_knowledge_base.md

@@ -128,7 +128,7 @@ See [Run retrieval test](./run_retrieval_test.md) for details.
 
 ## Search for knowledge base
 
-As of RAGFlow v0.20.0, the search feature is still in a rudimentary form, supporting only knowledge base search by name.
+As of RAGFlow v0.20.4, the search feature is still in a rudimentary form, supporting only knowledge base search by name.
 
 ![search knowledge base](https://github.com/infiniflow/ragflow/assets/93570324/836ae94c-2438-42be-879e-c7ad2a59693e)
 

docs/guides/manage_files.md

@@ -87,4 +87,4 @@ RAGFlow's file management allows you to download an uploaded file:
 
 ![download_file](https://github.com/infiniflow/ragflow/assets/93570324/cf3b297f-7d9b-4522-bf5f-4f45743e4ed5)
 
-> As of RAGFlow v0.20.0, bulk download is not supported, nor can you download an entire folder. 
+> As of RAGFlow v0.20.4, bulk download is not supported, nor can you download an entire folder. 

docs/guides/migration/migrate_from_docker_compose.md

@@ -0,0 +1,108 @@
+# Data Migration Guide
+
+A common scenario is processing large datasets on a powerful instance (e.g., with a GPU) and then migrating the entire RAGFlow service to a different production environment (e.g., a CPU-only server). This guide explains how to safely back up and restore your data using our provided migration script.
+
+## Identifying Your Data
+
+By default, RAGFlow uses Docker volumes to store all persistent data, including your database, uploaded files, and search indexes. You can see these volumes by running:
+
+```bash
+docker volume ls
+```
+
+The output will look similar to this:
+
+```text
+DRIVER    VOLUME NAME
+local     docker_esdata01
+local     docker_minio_data
+local     docker_mysql_data
+local     docker_redis_data
+```
+
+These volumes contain all the data you need to migrate.
+
+## Step 1: Stop RAGFlow Services
+
+Before starting the migration, you must stop all running RAGFlow services on the **source machine**. Navigate to the project's root directory and run:
+
+```bash
+docker-compose -f docker/docker-compose.yml down
+```
+
+**Important:** Do **not** use the `-v` flag (e.g., `docker-compose down -v`), as this will delete all your data volumes. The migration script includes a check and will prevent you from running it if services are active.
+
+## Step 2: Back Up Your Data
+
+We provide a convenient script to package all your data volumes into a single backup folder.
+
+For a quick reference of the script's commands and options, you can run:
+```bash
+bash docker/migration.sh help
+```
+
+To create a backup, run the following command from the project's root directory:
+
+```bash
+bash docker/migration.sh backup
+```
+
+This will create a `backup/` folder in your project root containing compressed archives of your data volumes.
+
+You can also specify a custom name for your backup folder:
+
+```bash
+bash docker/migration.sh backup my_ragflow_backup
+```
+
+This will create a folder named `my_ragflow_backup/` instead.
+
+## Step 3: Transfer the Backup Folder
+
+Copy the entire backup folder (e.g., `backup/` or `my_ragflow_backup/`) from your source machine to the RAGFlow project directory on your **target machine**. You can use tools like `scp`, `rsync`, or a physical drive for the transfer.
+
+## Step 4: Restore Your Data
+
+On the **target machine**, ensure that RAGFlow services are not running. Then, use the migration script to restore your data from the backup folder.
+
+If your backup folder is named `backup/`, run:
+
+```bash
+bash docker/migration.sh restore
+```
+
+If you used a custom name, specify it in the command:
+
+```bash
+bash docker/migration.sh restore my_ragflow_backup
+```
+
+The script will automatically create the necessary Docker volumes and unpack the data.
+
+**Note:** If the script detects that Docker volumes with the same names already exist on the target machine, it will warn you that restoring will overwrite the existing data and ask for confirmation before proceeding.
+
+## Step 5: Start RAGFlow Services
+
+Once the restore process is complete, you can start the RAGFlow services on your new machine:
+
+```bash
+docker-compose -f docker/docker-compose.yml up -d
+```
+
+**Note:** If you already have build an service by docker-compose before, you may need to backup your data for target machine like this guide above and run like:
+
+```bash
+# Please backup by `sh docker/migration.sh backup backup_dir_name` before you do the following line.
+# !!! this line -v flag will delete the original docker volume
+docker-compose -f docker/docker-compose.yml down -v
+docker-compose -f docker/docker-compose.yml up -d
+```
+
+Your RAGFlow instance is now running with all the data from your original machine.
+
+
+
+
+
+
+

docs/guides/tracing.mdx

@@ -18,7 +18,7 @@ RAGFlow ships with a built-in [Langfuse](https://langfuse.com) integration so th
 Langfuse stores traces, spans and prompt payloads in a purpose-built observability backend and offers filtering and visualisations on top.  
 
 :::info NOTE
-• RAGFlow **≥ 0.20.0** (contains the Langfuse connector)  
+• RAGFlow **≥ 0.20.4** (contains the Langfuse connector)  
 • A Langfuse workspace (cloud or self-hosted) with a _Project Public Key_ and _Secret Key_
 :::
 

docs/guides/upgrade_ragflow.mdx

@@ -66,10 +66,10 @@ To upgrade RAGFlow, you must upgrade **both** your code **and** your Docker imag
    git clone https://github.com/infiniflow/ragflow.git
    ```
 
-2. Switch to the latest, officially published release, e.g., `v0.20.0`:
+2. Switch to the latest, officially published release, e.g., `v0.20.4`:
 
    ```bash
-   git checkout -f v0.20.0
+   git checkout -f v0.20.4
    ```
 
 3. Update **ragflow/docker/.env**:
@@ -83,14 +83,14 @@ To upgrade RAGFlow, you must upgrade **both** your code **and** your Docker imag
   <TabItem value="slim">
 
 ```bash
-RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.0-slim
+RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.4-slim
 ```
 
   </TabItem>
   <TabItem value="full">
 
 ```bash
-RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.0
+RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.4
 ```
 
   </TabItem>
@@ -114,10 +114,10 @@ No, you do not need to. Upgrading RAGFlow in itself will *not* remove your uploa
 1. From an environment with Internet access, pull the required Docker image.
 2. Save the Docker image to a **.tar** file.
    ```bash
-   docker save -o ragflow.v0.20.0.tar infiniflow/ragflow:v0.20.0
+   docker save -o ragflow.v0.20.4.tar infiniflow/ragflow:v0.20.4
    ```
 3. Copy the **.tar** file to the target server.
 4. Load the **.tar** file into Docker:
    ```bash
-   docker load -i ragflow.v0.20.0.tar
+   docker load -i ragflow.v0.20.4.tar
    ```

docs/quickstart.mdx

@@ -44,7 +44,7 @@ This section provides instructions on setting up the RAGFlow server on Linux. If
 
    `vm.max_map_count`. This value sets the maximum number of memory map areas a process may have. Its default value is 65530. While most applications require fewer than a thousand maps, reducing this value can result in abnormal behaviors, and the system will throw out-of-memory errors when a process reaches the limitation.
 
-   RAGFlow v0.20.0 uses Elasticsearch or [Infinity](https://github.com/infiniflow/infinity) for multiple recall. Setting the value of `vm.max_map_count` correctly is crucial to the proper functioning of the Elasticsearch component.
+   RAGFlow v0.20.4 uses Elasticsearch or [Infinity](https://github.com/infiniflow/infinity) for multiple recall. Setting the value of `vm.max_map_count` correctly is crucial to the proper functioning of the Elasticsearch component.
 
 <Tabs
   defaultValue="linux"
@@ -184,13 +184,13 @@ This section provides instructions on setting up the RAGFlow server on Linux. If
    ```bash
    $ git clone https://github.com/infiniflow/ragflow.git
    $ cd ragflow/docker
-   $ git checkout -f v0.20.0
+   $ git checkout -f v0.20.4
    ```
 
 3. Use the pre-built Docker images and start up the server:
 
    :::tip NOTE
-   The command below downloads the `v0.20.0-slim` edition of the RAGFlow Docker image. Refer to the following table for descriptions of different RAGFlow editions. To download a RAGFlow edition different from `v0.20.0-slim`, update the `RAGFLOW_IMAGE` variable accordingly in **docker/.env** before using `docker compose` to start the server. For example: set `RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.0` for the full edition `v0.20.0`.
+   The command below downloads the `v0.20.4-slim` edition of the RAGFlow Docker image. Refer to the following table for descriptions of different RAGFlow editions. To download a RAGFlow edition different from `v0.20.4-slim`, update the `RAGFLOW_IMAGE` variable accordingly in **docker/.env** before using `docker compose` to start the server. For example: set `RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.4` for the full edition `v0.20.4`.
    :::
 
    ```bash
@@ -207,8 +207,8 @@ This section provides instructions on setting up the RAGFlow server on Linux. If
 
 | RAGFlow image tag   | Image size (GB) | Has embedding models and Python packages? | Stable?                  |
 | ------------------- | --------------- | ----------------------------------------- | ------------------------ |
-| `v0.20.0`           | &approx;9       | :heavy_check_mark:                        | Stable release           |
-| `v0.20.0-slim`      | &approx;2       | ❌                                        | Stable release           |
+| `v0.20.4`           | &approx;9       | :heavy_check_mark:                        | Stable release           |
+| `v0.20.4-slim`      | &approx;2       | ❌                                        | Stable release           |
 | `nightly`           | &approx;9       | :heavy_check_mark:                        | *Unstable* nightly build |
 | `nightly-slim`      | &approx;2       | ❌                                        | *Unstable* nightly build |
 
@@ -217,7 +217,7 @@ This section provides instructions on setting up the RAGFlow server on Linux. If
 ```
 
 :::danger IMPORTANT
-The embedding models included in `v0.20.0` and `nightly` are:
+The embedding models included in `v0.20.4` and `nightly` are:
 
 - BAAI/bge-large-zh-v1.5
 - maidalun1020/bce-embedding-base_v1

docs/references/glossary.mdx

@@ -19,7 +19,7 @@ import TOCInline from '@theme/TOCInline';
 
 ### Cross-language search
 
-Cross-language search (also known as cross-lingual retrieval) is a feature introduced in version 0.20.0. It enables users to submit queries in one language (for example, English) and retrieve relevant documents written in other languages such as Chinese or Spanish. This feature is enabled by the system’s default chat model, which translates queries to ensure accurate matching of semantic meaning across languages.
+Cross-language search (also known as cross-lingual retrieval) is a feature introduced in version 0.20.4. It enables users to submit queries in one language (for example, English) and retrieve relevant documents written in other languages such as Chinese or Spanish. This feature is enabled by the system’s default chat model, which translates queries to ensure accurate matching of semantic meaning across languages.
 
 By enabling cross-language search, users can effortlessly access a broader range of information regardless of language barriers, significantly enhancing the system’s usability and inclusiveness.
 

docs/references/python_api_reference.md

@@ -5,7 +5,7 @@ slug: /python_api_reference
 
 # Python API
 
-A complete reference for RAGFlow's Python APIs. Before proceeding, please ensure you [have your RAGFlow API key ready for authentication](../guides/models/llm_api_key_setup.md).
+A complete reference for RAGFlow's Python APIs. Before proceeding, please ensure you [have your RAGFlow API key ready for authentication](https://ragflow.io/docs/dev/acquire_ragflow_api_key).
 
 :::tip NOTE
 Run the following command to download the Python SDK:
@@ -507,7 +507,16 @@ print(doc)
 ### List documents
 
 ```python
-Dataset.list_documents(id:str =None, keywords: str=None, page: int=1, page_size:int = 30, order_by:str = "create_time", desc: bool = True) -> list[Document]
+Dataset.list_documents(
+    id: str = None,
+    keywords: str = None,
+    page: int = 1,
+    page_size: int = 30,
+    order_by: str = "create_time",
+    desc: bool = True,
+    create_time_from: int = 0,
+    create_time_to: int = 0
+) -> list[Document]
 ```
 
 Lists documents in the current dataset.
@@ -541,6 +550,12 @@ The field by which documents should be sorted. Available options:
 
 Indicates whether the retrieved documents should be sorted in descending order. Defaults to `True`.
 
+##### create_time_from: `int`
+Unix timestamp for filtering documents created after this time. 0 means no filter. Defaults to 0.
+
+##### create_time_to: `int`
+Unix timestamp for filtering documents created before this time. 0 means no filter. Defaults to 0.
+
 #### Returns
 
 - Success: A list of `Document` objects.

docs/release_notes.md

@@ -9,8 +9,8 @@ Key features, improvements and bug fixes in the latest releases.
 
 :::info
 Each RAGFlow release is available in two editions:
-- **Slim edition**: excludes built-in embedding models and is identified by a **-slim** suffix added to the version name. Example: `infiniflow/ragflow:v0.19.1-slim`
-- **Full edition**: includes built-in embedding models and has no suffix added to the version name. Example: `infiniflow/ragflow:v0.19.1`
+- **Slim edition**: excludes built-in embedding models and is identified by a **-slim** suffix added to the version name. Example: `infiniflow/ragflow:v0.20.4-slim`
+- **Full edition**: includes built-in embedding models and has no suffix added to the version name. Example: `infiniflow/ragflow:v0.20.4`
 :::
 
 :::danger IMPORTANT
@@ -22,6 +22,129 @@ The embedding models included in a full edition are:
 These two embedding models are optimized specifically for English and Chinese, so performance may be compromised if you use them to embed documents in other languages.
 :::
 
+## v0.20.4
+
+Released on August 27, 2025.
+
+### Improvements
+
+- Agent component: Completes Chinese localization for the Agent component.
+- Introduces the `ENABLE_TIMEOUT_ASSERTION` environment variable to enable or disable timeout assertions for file parsing tasks.
+- Dataset:
+  - Improves Markdown file parsing, with AST support to avoid unintended chunking.
+  - Enhances HTML parsing, supporting bs4-based HTML tag traversal.
+
+### Added models
+
+ZHIPU GLM-4.5
+
+### New Agent templates
+
+Ecommerce Customer Service Workflow: A template designed to handle enquiries about product features and multi-product comparisons using the internal knowledge base, as well as to manage installation appointment bookings.
+
+### Fixed issues
+
+- Dataset:  
+  - Unable to share resources with the team.
+  - Inappropriate restrictions on the number and size of uploaded files.
+- Chat:
+  - Unable to preview referenced files in responses.
+  - Unable to send out messages after file uploads.
+- An OAuth2 authentication failure.
+- A logical error in multi-conditioned metadata searches within a dataset.
+- Citations infinitely increased in multi-turn conversations.
+
+## v0.20.3
+
+Released on August 20, 2025.
+
+### Improvements
+
+- Revamps the user interface for the **Datasets**, **Chat**, and **Search** pages.  
+- Search and Chat: Introduces document-level metadata filtering, allowing automatic or manual filtering during chats or searches.
+- Search: Supports creating search apps tailored to various business scenarios
+- Chat: Supports comparing answer performance of up to three chat model settings on a single **Chat** page.
+- Agent:  
+  - Implements a toggle in the **Agent** component to enable or disable citation.  
+  - Introduces a drag-and-drop method for creating components.  
+- Documentation: Corrects inaccuracies in the API reference.
+
+### New Agent templates
+
+- Report Agent: A template for generating summary reports in internal question-answering scenarios, supporting the display of tables and formulae.  [#9427](https://github.com/infiniflow/ragflow/pull/9427)
+
+### Fixed issues
+
+- The timeout mechanism introduced in v0.20.0 caused tasks like GraphRAG to halt.
+- Predefined opening greeting in the **Agent** component was missing during conversations.  
+- An automatic line break issue in the prompt editor.  
+- A memory leak issue caused by PyPDF. [#9469](https://github.com/infiniflow/ragflow/pull/9469)
+
+### API changes
+
+#### Deprecated
+
+[Create session with agent](./references/http_api_reference.md#create-session-with-agent)
+
+## v0.20.1
+
+Released on August 8, 2025.
+
+### New Features
+
+- The **Retrieval** component now supports the dynamic specification of knowledge base names using variables.
+- The user interface now includes a French language option.
+
+### Added Models
+
+- GPT-5
+- Claude 4.1
+
+### New agent templates (both workflow and agentic)
+
+- SQL Assistant Workflow: Empowers non-technical teams (e.g., operations, product) to independently query business data.
+- Choose Your Knowledge Base Workflow: Lets users select a knowledge base to query during conversations. [#9325](https://github.com/infiniflow/ragflow/pull/9325)
+- Choose Your Knowledge Base Agent: Delivers higher-quality responses with extended reasoning time, suited for complex queries. [#9325](https://github.com/infiniflow/ragflow/pull/9325)
+
+### Fixed Issues
+
+- The **Agent** component was unable to invoke models installed via vLLM.
+- Agents could not be shared with the team.
+- Embedding an Agent into a webpage was not functioning properly.
+
+## v0.20.0
+
+Released on August 4, 2025.
+
+### Compatibility changes
+
+From v0.20.0 onwards, Agents are no longer compatible with earlier versions, and all existing Agents from previous versions must be rebuilt following the upgrade.
+
+### New features
+
+- Unified orchestration of both Agents and Workflows.
+- A comprehensive refactor of the Agent, greatly enhancing its capabilities and usability, with support for Multi-Agent configurations, planning and reflection, and visual functionalities.
+- Fully implemented MCP functionality, allowing for MCP Server import, Agents functioning as MCP Clients, and RAGFlow itself operating as an MCP Server.
+- Access to runtime logs for Agents.
+- Chat histories with Agents available through the management panel.
+- Integration of a new, more robust version of Infinity, enabling the auto-tagging functionality with Infinity as the underlying document engine.
+- An OpenAI-compatible API that supports file reference information.
+- Support for new models, including Kimi K2, Grok 4, and Voyage embedding.
+- RAGFlow’s codebase is now mirrored on Gitee.
+- Introduction of a new model provider, Gitee AI.
+
+### New agent templates introduced
+
+- Multi-Agent based Deep Research: Collaborative Agent teamwork led by a Lead Agent with multiple Subagents, distinct from traditional workflow orchestration.
+- An intelligent Q&A chatbot leveraging internal knowledge bases, designed for customer service and training scenarios.
+- A resume analysis template used by the RAGFlow team to screen, analyze, and record candidate information.
+- A blog generation workflow that transforms raw ideas into SEO-friendly blog content.
+- An intelligent customer service workflow.
+- A user feedback analysis template that directs user feedback to appropriate teams through semantic analysis.
+- Trip Planner: Uses web search and map MCP servers to assist with travel planning.
+- Image Lingo: Translates content from uploaded photos.
+- An information search assistant that retrieves answers from both internal knowledge bases and the web.
+
 ## v0.19.1
 
 Released on June 23, 2025.
@@ -123,7 +246,7 @@ From this release onwards, if you still see RAGFlow's responses being cut short
 
 - Unable to add models via Ollama/Xinference, an issue introduced in v0.17.1.
 
-### Related APIs
+### API changes
 
 #### HTTP APIs
 
@@ -184,7 +307,7 @@ The following is a screenshot of a conversation that integrates Deep Research:
 
 ![Image](https://github.com/user-attachments/assets/165b88ff-1f5d-4fb8-90e2-c836b25e32e9)
 
-### Related APIs
+### API changes
 
 #### HTTP APIs
 
@@ -259,7 +382,7 @@ This release fixes the following issues:
 - Using the **Table** parsing method results in information loss.
 - Miscellaneous API issues.
 
-### Related APIs
+### API changes
 
 #### HTTP APIs
 
@@ -295,7 +418,7 @@ Released on December 18, 2024.
 - Upgrades the Document Layout Analysis model in DeepDoc.
 - Significantly enhances the retrieval performance when using [Infinity](https://github.com/infiniflow/infinity) as document engine.
 
-### Related APIs
+### API changes
 
 #### HTTP APIs
 
@@ -352,7 +475,7 @@ This approach eliminates the need to manually update **service_config.yaml** aft
 Ensure that you [upgrade **both** your code **and** Docker image to this release](https://ragflow.io/docs/dev/upgrade_ragflow#upgrade-ragflow-to-the-most-recent-officially-published-release) before trying this new approach.
 :::
 
-### Related APIs
+### API changes
 
 #### HTTP APIs
 
@@ -511,7 +634,7 @@ While we also test RAGFlow on ARM64 platforms, we do not maintain RAGFlow Docker
 If you are on an ARM platform, follow [this guide](./develop/build_docker_image.mdx) to build a RAGFlow Docker image.
 :::
 
-### Related APIs
+### API changes
 
 #### HTTP API
 
@@ -532,7 +655,7 @@ Released on May 21, 2024.
 - Supports monitoring of system components, including Elasticsearch, MySQL, Redis, and MinIO.
 - Supports disabling **Layout Recognition** in the GENERAL chunking method to reduce file chunking time.
 
-### Related APIs
+### API changes
 
 #### HTTP API
 

graphrag/entity_resolution.py

@@ -15,6 +15,7 @@
 #
 import logging
 import itertools
+import os
 import re
 from dataclasses import dataclass
 from typing import Any, Callable
@@ -106,7 +107,8 @@ class EntityResolution(Extractor):
             nonlocal remain_candidates_to_resolve, callback
             async with semaphore:
                 try:
-                    with trio.move_on_after(180) as cancel_scope:
+                    enable_timeout_assertion = os.environ.get("ENABLE_TIMEOUT_ASSERTION")
+                    with trio.move_on_after(280 if enable_timeout_assertion else 1000000000) as cancel_scope:
                         await self._resolve_candidate(candidate_batch, result_set, result_lock)
                         remain_candidates_to_resolve = remain_candidates_to_resolve - len(candidate_batch[1])
                         callback(msg=f"Resolved {len(candidate_batch[1])} pairs, {remain_candidates_to_resolve} are remained to resolve. ")
@@ -169,7 +171,8 @@ class EntityResolution(Extractor):
         logging.info(f"Created resolution prompt {len(text)} bytes for {len(candidate_resolution_i[1])} entity pairs of type {candidate_resolution_i[0]}")
         async with chat_limiter:
             try:
-                with trio.move_on_after(120) as cancel_scope:
+                enable_timeout_assertion = os.environ.get("ENABLE_TIMEOUT_ASSERTION")
+                with trio.move_on_after(280 if enable_timeout_assertion else 1000000000) as cancel_scope:
                     response = await trio.to_thread.run_sync(self._chat, text, [{"role": "user", "content": "Output:"}], {})
                 if cancel_scope.cancelled_caught:
                     logging.warning("_resolve_candidate._chat timeout, skipping...")

graphrag/general/community_reports_extractor.py

@@ -7,6 +7,7 @@ Reference:
 
 import logging
 import json
+import os
 import re
 from typing import Callable
 from dataclasses import dataclass
@@ -51,6 +52,7 @@ class CommunityReportsExtractor(Extractor):
         self._max_report_length = max_report_length or 1500
 
     async def __call__(self, graph: nx.Graph, callback: Callable | None = None):
+        enable_timeout_assertion = os.environ.get("ENABLE_TIMEOUT_ASSERTION")
         for node_degree in graph.degree:
             graph.nodes[str(node_degree[0])]["rank"] = int(node_degree[1])
 
@@ -92,7 +94,7 @@ class CommunityReportsExtractor(Extractor):
             text = perform_variable_replacements(self._extraction_prompt, variables=prompt_variables)
             async with chat_limiter:
                 try:
-                    with trio.move_on_after(80) as cancel_scope:
+                    with trio.move_on_after(180 if enable_timeout_assertion else 1000000000) as cancel_scope:
                         response = await trio.to_thread.run_sync( self._chat, text, [{"role": "user", "content": "Output:"}], {})
                     if cancel_scope.cancelled_caught:
                         logging.warning("extract_community_report._chat timeout, skipping...")

graphrag/general/extractor.py

@@ -47,7 +47,7 @@ class Extractor:
         self._language = language
         self._entity_types = entity_types or DEFAULT_ENTITY_TYPES
 
-    @timeout(60*3)
+    @timeout(60*20)
     def _chat(self, system, history, gen_conf={}):
         hist = deepcopy(history)
         conf = deepcopy(gen_conf)

graphrag/general/index.py

@@ -15,6 +15,8 @@
 #
 import json
 import logging
+import os
+
 import networkx as nx
 import trio
 
@@ -49,6 +51,7 @@ async def run_graphrag(
     embedding_model,
     callback,
 ):
+    enable_timeout_assertion=os.environ.get("ENABLE_TIMEOUT_ASSERTION")
     start = trio.current_time()
     tenant_id, kb_id, doc_id = row["tenant_id"], str(row["kb_id"]), row["doc_id"]
     chunks = []
@@ -57,20 +60,22 @@ async def run_graphrag(
     ):
         chunks.append(d["content_with_weight"])
 
-    subgraph = await generate_subgraph(
-        LightKGExt
-        if "method" not in row["kb_parser_config"].get("graphrag", {}) or row["kb_parser_config"]["graphrag"]["method"] != "general"
-        else GeneralKGExt,
-        tenant_id,
-        kb_id,
-        doc_id,
-        chunks,
-        language,
-        row["kb_parser_config"]["graphrag"].get("entity_types", []),
-        chat_model,
-        embedding_model,
-        callback,
-    )
+    with trio.fail_after(max(120, len(chunks)*60*10) if enable_timeout_assertion else 10000000000):
+        subgraph = await generate_subgraph(
+            LightKGExt
+            if "method" not in row["kb_parser_config"].get("graphrag", {}) or row["kb_parser_config"]["graphrag"]["method"] != "general"
+            else GeneralKGExt,
+            tenant_id,
+            kb_id,
+            doc_id,
+            chunks,
+            language,
+            row["kb_parser_config"]["graphrag"].get("entity_types", []),
+            chat_model,
+            embedding_model,
+            callback,
+        )
+
     if not subgraph:
         return
 
@@ -125,7 +130,6 @@ async def run_graphrag(
     return
 
 
-@timeout(60*60, 1)
 async def generate_subgraph(
     extractor: Extractor,
     tenant_id: str,

graphrag/light/graph_prompt.py

@@ -130,7 +130,36 @@ Output:
 
 PROMPTS[
     "entiti_continue_extraction"
-] = """MANY entities were missed in the last extraction.  Add them below using the same format:
+] = """
+MANY entities and relationships were missed in the last extraction. Please find only the missing entities and relationships from previous text.
+
+---Remember Steps---
+
+1. Identify all entities. For each identified entity, extract the following information:
+- entity_name: Name of the entity, use same language as input text. If English, capitalized the name
+- entity_type: One of the following types: [{entity_types}]
+- entity_description: Provide a comprehensive description of the entity's attributes and activities *based solely on the information present in the input text*. **Do not infer or hallucinate information not explicitly stated.** If the text provides insufficient information to create a comprehensive description, state "Description not available in text."
+Format each entity as ("entity"{tuple_delimiter}<entity_name>{tuple_delimiter}<entity_type>{tuple_delimiter}<entity_description>)
+
+2. From the entities identified in step 1, identify all pairs of (source_entity, target_entity) that are *clearly related* to each other.
+For each pair of related entities, extract the following information:
+- source_entity: name of the source entity, as identified in step 1
+- target_entity: name of the target entity, as identified in step 1
+- relationship_description: explanation as to why you think the source entity and the target entity are related to each other
+- relationship_strength: a numeric score indicating strength of the relationship between the source entity and target entity
+- relationship_keywords: one or more high-level key words that summarize the overarching nature of the relationship, focusing on concepts or themes rather than specific details
+Format each relationship as ("relationship"{tuple_delimiter}<source_entity>{tuple_delimiter}<target_entity>{tuple_delimiter}<relationship_description>{tuple_delimiter}<relationship_keywords>{tuple_delimiter}<relationship_strength>)
+
+3. Identify high-level key words that summarize the main concepts, themes, or topics of the entire text. These should capture the overarching ideas present in the document.
+Format the content-level key words as ("content_keywords"{tuple_delimiter}<high_level_keywords>)
+
+4. Return output in {language} as a single list of all the entities and relationships identified in steps 1 and 2. Use **{record_delimiter}** as the list delimiter.
+
+5. When finished, output {completion_delimiter}
+
+---Output---
+
+Add new entities and relations below using the same format, and do not include entities and relations that have been previously extracted. :
 """
 
 PROMPTS[
@@ -252,4 +281,4 @@ When handling information with timestamps:
 - List up to 5 most important reference sources at the end under "References", clearly indicating whether each source is from Knowledge Graph (KG) or Vector Data (VD)
   Format: [KG/VD] Source content
 
-Add sections and commentary to the response as appropriate for the length and format. If the provided information is insufficient to answer the question, clearly state that you don't know or cannot provide an answer in the same language as the user's question."""
+Add sections and commentary to the response as appropriate for the length and format. If the provided information is insufficient to answer the question, clearly state that you don't know or cannot provide an answer in the same language as the user's question."""

graphrag/utils.py

@@ -307,6 +307,7 @@ def chunk_id(chunk):
 
 async def graph_node_to_chunk(kb_id, embd_mdl, ent_name, meta, chunks):
     global chat_limiter
+    enable_timeout_assertion=os.environ.get("ENABLE_TIMEOUT_ASSERTION")
     chunk = {
         "id": get_uuid(),
         "important_kwd": [ent_name],
@@ -324,7 +325,7 @@ async def graph_node_to_chunk(kb_id, embd_mdl, ent_name, meta, chunks):
     ebd = get_embed_cache(embd_mdl.llm_name, ent_name)
     if ebd is None:
         async with chat_limiter:
-            with trio.fail_after(3):
+            with trio.fail_after(3 if enable_timeout_assertion else 30000000):
                 ebd, _ = await trio.to_thread.run_sync(lambda: embd_mdl.encode([ent_name]))
         ebd = ebd[0]
         set_embed_cache(embd_mdl.llm_name, ent_name, ebd)
@@ -362,6 +363,7 @@ def get_relation(tenant_id, kb_id, from_ent_name, to_ent_name, size=1):
 
 
 async def graph_edge_to_chunk(kb_id, embd_mdl, from_ent_name, to_ent_name, meta, chunks):
+    enable_timeout_assertion=os.environ.get("ENABLE_TIMEOUT_ASSERTION")
     chunk = {
         "id": get_uuid(),
         "from_entity_kwd": from_ent_name,
@@ -380,7 +382,7 @@ async def graph_edge_to_chunk(kb_id, embd_mdl, from_ent_name, to_ent_name, meta,
     ebd = get_embed_cache(embd_mdl.llm_name, txt)
     if ebd is None:
         async with chat_limiter:
-            with trio.fail_after(3):
+            with trio.fail_after(3 if enable_timeout_assertion else 300000000):
                 ebd, _ = await trio.to_thread.run_sync(lambda: embd_mdl.encode([txt+f": {meta['description']}"]))
         ebd = ebd[0]
         set_embed_cache(embd_mdl.llm_name, txt, ebd)
@@ -514,9 +516,10 @@ async def set_graph(tenant_id: str, kb_id: str, embd_mdl, graph: nx.Graph, chang
         callback(msg=f"set_graph converted graph change to {len(chunks)} chunks in {now - start:.2f}s.")
     start = now
 
+    enable_timeout_assertion=os.environ.get("ENABLE_TIMEOUT_ASSERTION")
     es_bulk_size = 4
     for b in range(0, len(chunks), es_bulk_size):
-        with trio.fail_after(3):
+        with trio.fail_after(3 if enable_timeout_assertion else 30000000):
             doc_store_result = await trio.to_thread.run_sync(lambda: settings.docStoreConn.insert(chunks[b:b + es_bulk_size], search.index_name(tenant_id), kb_id))
         if b % 100 == es_bulk_size and callback:
             callback(msg=f"Insert chunks: {b}/{len(chunks)}")

helm/templates/elasticsearch.yaml

@@ -44,9 +44,21 @@ spec:
         checksum/config-es: {{ include (print $.Template.BasePath "/elasticsearch-config.yaml") . | sha256sum }}
         checksum/config-env: {{ include (print $.Template.BasePath "/env.yaml") . | sha256sum }}
     spec:
+      {{- if or .Values.imagePullSecrets .Values.elasticsearch.image.pullSecrets }}
+      imagePullSecrets:
+        {{- with .Values.imagePullSecrets }}
+        {{- toYaml . | nindent 8 }}
+        {{- end }}
+        {{- with .Values.elasticsearch.image.pullSecrets }}
+        {{- toYaml . | nindent 8 }}
+        {{- end }}
+      {{- end }}
       initContainers:
       - name: fix-data-volume-permissions
-        image: alpine
+        image: {{ .Values.elasticsearch.initContainers.alpine.repository }}:{{ .Values.elasticsearch.initContainers.alpine.tag }}
+        {{- with .Values.elasticsearch.initContainers.alpine.pullPolicy }}
+        imagePullPolicy: {{ . }}
+        {{- end }}
         command:
         - sh
         - -c
@@ -55,14 +67,20 @@ spec:
           - mountPath: /usr/share/elasticsearch/data
             name: es-data
       - name: sysctl
-        image: busybox
+        image: {{ .Values.elasticsearch.initContainers.busybox.repository }}:{{ .Values.elasticsearch.initContainers.busybox.tag }}
+        {{- with .Values.elasticsearch.initContainers.busybox.pullPolicy }}
+        imagePullPolicy: {{ . }}
+        {{- end }}
         securityContext:
           privileged: true
           runAsUser: 0
         command: ["sysctl", "-w", "vm.max_map_count=262144"]
       containers:
       - name: elasticsearch
-        image: elasticsearch:{{ .Values.env.STACK_VERSION }}
+        image: {{ .Values.elasticsearch.image.repository }}:{{ .Values.elasticsearch.image.tag }}
+        {{- with .Values.elasticsearch.image.pullPolicy }}
+        imagePullPolicy: {{ . }}
+        {{- end }}
         envFrom:
           - secretRef:
               name: {{ include "ragflow.fullname" . }}-env-config

helm/templates/infinity.yaml

@@ -43,9 +43,21 @@ spec:
       annotations:
         checksum/config: {{ include (print $.Template.BasePath "/env.yaml") . | sha256sum }}
     spec:
+      {{- if or .Values.imagePullSecrets .Values.infinity.image.pullSecrets }}
+      imagePullSecrets:
+        {{- with .Values.imagePullSecrets }}
+        {{- toYaml . | nindent 8 }}
+        {{- end }}
+        {{- with .Values.infinity.image.pullSecrets }}
+        {{- toYaml . | nindent 8 }}
+        {{- end }}
+      {{- end }}
       containers:
       - name: infinity
         image: {{ .Values.infinity.image.repository }}:{{ .Values.infinity.image.tag }}
+        {{- with .Values.infinity.image.pullPolicy }}
+        imagePullPolicy: {{ . }}
+        {{- end }}
         envFrom:
           - secretRef:
               name: {{ include "ragflow.fullname" . }}-env-config

helm/templates/minio.yaml

@@ -43,9 +43,21 @@ spec:
         {{- include "ragflow.labels" . | nindent 8 }}
         app.kubernetes.io/component: minio
     spec:
+      {{- if or .Values.imagePullSecrets .Values.minio.image.pullSecrets }}
+      imagePullSecrets:
+        {{- with .Values.imagePullSecrets }}
+        {{- toYaml . | nindent 8 }}
+        {{- end }}
+        {{- with .Values.minio.image.pullSecrets }}
+        {{- toYaml . | nindent 8 }}
+        {{- end }}
+      {{- end }}
       containers:
       - name: minio
         image: {{ .Values.minio.image.repository }}:{{ .Values.minio.image.tag }}
+        {{- with .Values.minio.image.pullPolicy }}
+        imagePullPolicy: {{ . }}
+        {{- end }}
         envFrom:
           - secretRef:
               name: {{ include "ragflow.fullname" . }}-env-config