Use 'float' explicitly for OpenAI's embedding "encoding_format" (#9838)

### What problem does this PR solve?

The default value for OpenAI '/v1/embeddings' parameter
'encoding_format' is 'base64'. Use 'float' explicitly to avoid base64
encoding & decoding, larger data size.


https://github.com/openai/openai-python/blob/main/src/openai/resources/embeddings.py
        if not is_given(encoding_format):
            params["encoding_format"] = "base64"

### Type of change

- [x] Performance Improvement

.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

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:v0.20.1">
+        <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">
@@ -190,7 +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 `v0.20.1-slim` edition of the RAGFlow Docker image. See the following table for descriptions of different RAGFlow editions. To download a RAGFlow edition different from `v0.20.1-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:v0.20.1` for the full edition `v0.20.1`.
+   > 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
@@ -203,8 +203,8 @@ releases! 🌟
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    |-------------------|-----------------|-----------------------|--------------------------|
-   | v0.20.1           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.20.1-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  |
 
@@ -307,7 +307,7 @@ docker build --platform linux/amd64 -f Dockerfile -t infiniflow/ragflow:nightly
 
 ## 🔨 Launch service from source for development
 
-1. Install uv, or skip this step if it is already installed:
+1. Install `uv` and `pre-commit`, or skip this step if they are already installed:
 
    ```bash
    pipx install uv pre-commit

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:v0.20.1">
+        <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">
@@ -181,7 +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 v0.20.1-slim dari gambar Docker RAGFlow. Silakan merujuk ke tabel berikut untuk deskripsi berbagai edisi RAGFlow. Untuk mengunduh edisi RAGFlow yang berbeda dari v0.20.1-slim, perbarui variabel RAGFLOW_IMAGE di docker/.env sebelum menggunakan docker compose untuk memulai server. Misalnya, atur RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.1 untuk edisi lengkap v0.20.1.
+> 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
@@ -194,8 +194,8 @@ $ docker compose -f docker-compose.yml up -d
 
 | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
 | ----------------- | --------------- | --------------------- | ------------------------ |
-| v0.20.1           | &approx;9       | :heavy_check_mark:    | Stable release           |
-| v0.20.1-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 |
 
@@ -271,7 +271,7 @@ docker build --platform linux/amd64 -f Dockerfile -t infiniflow/ragflow:nightly
 
 ## 🔨 Menjalankan Aplikasi dari untuk Pengembangan
 
-1. Instal uv, atau lewati langkah ini jika sudah terinstal:
+1. Instal `uv` dan `pre-commit`, atau lewati langkah ini jika sudah terinstal:
 
    ```bash
    pipx install uv pre-commit

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:v0.20.1">
+        <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">
@@ -160,7 +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 v0.20.1-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。v0.20.1-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 v0.20.1 をダウンロードするには、RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.1 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
@@ -173,8 +173,8 @@
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    | ----------------- | --------------- | --------------------- | ------------------------ |
-   | v0.20.1           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.20.1-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 |
 
@@ -266,7 +266,7 @@ docker build --platform linux/amd64 -f Dockerfile -t infiniflow/ragflow:nightly
 
 ## 🔨 ソースコードからサービスを起動する方法
 
-1. uv をインストールする。すでにインストールされている場合は、このステップをスキップしてください:
+1. `uv` と `pre-commit` をインストールする。すでにインストールされている場合は、このステップをスキップしてください:
 
    ```bash
    pipx install uv pre-commit

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:v0.20.1">
+        <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">
@@ -160,7 +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 v0.20.1-slim 버전을 다운로드합니다. 다양한 RAGFlow 버전에 대한 설명은 다음 표를 참조하십시오. v0.20.1-slim과 다른 RAGFlow 버전을 다운로드하려면, docker/.env 파일에서 RAGFLOW_IMAGE 변수를 적절히 업데이트한 후 docker compose를 사용하여 서버를 시작하십시오. 예를 들어, 전체 버전인 v0.20.1을 다운로드하려면 RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.1로 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
@@ -173,8 +173,8 @@
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    | ----------------- | --------------- | --------------------- | ------------------------ |
-   | v0.20.1           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.20.1-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 |
 
@@ -265,7 +265,7 @@ docker build --platform linux/amd64 -f Dockerfile -t infiniflow/ragflow:nightly
 
 ## 🔨 소스 코드로 서비스를 시작합니다.
 
-1. uv를 설치하거나 이미 설치된 경우 이 단계를 건너뜁니다:
+1. `uv` 와 `pre-commit` 을 설치하거나, 이미 설치된 경우 이 단계를 건너뜁니다:
 
    ```bash
    pipx install uv pre-commit

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:v0.20.1">
+        <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">
@@ -180,7 +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 `v0.20.1-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 `v0.20.1-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:v0.20.1` para a edição completa `v0.20.1`.
+    > 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
@@ -193,8 +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?                 |
     | --------------------- | ---------------------- | ------------------------------- | ------------------------ |
-    | v0.20.1               | ~9                     | :heavy_check_mark:              | Lançamento estável       |
-    | v0.20.1-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 |
 
@@ -289,7 +289,7 @@ docker build --platform linux/amd64 -f Dockerfile -t infiniflow/ragflow:nightly
 
 ## 🔨 Lançar o serviço a partir do código-fonte para desenvolvimento
 
-1. Instale o `uv`, ou pule esta etapa se ele já estiver instalado:
+1. Instale o `uv` e o `pre-commit`, ou pule esta etapa se eles já estiverem instalados:
 
    ```bash
    pipx install uv pre-commit

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:v0.20.1">
+        <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">
@@ -183,7 +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 `v0.20.1-slim`。請參考下表查看不同 Docker 發行版的說明。如需下載不同於 `v0.20.1-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:v0.20.1` 來下載 RAGFlow 鏡像的 `v0.20.1` 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
@@ -196,8 +196,8 @@
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    | ----------------- | --------------- | --------------------- | ------------------------ |
-   | v0.20.1           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.20.1-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 |
 
@@ -301,7 +301,7 @@ docker build --platform linux/amd64 --build-arg NEED_MIRROR=1 -f Dockerfile -t i
 
 ## 🔨 以原始碼啟動服務
 
-1. 安裝 uv。如已安裝，可跳過此步驟：
+1. 安裝 `uv` 和 `pre-commit`。如已安裝，可跳過此步驟：
 
    ```bash
    pipx install uv pre-commit

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:v0.20.1">
+        <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">
@@ -183,7 +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 `v0.20.1-slim`。请参考下表查看不同 Docker 发行版的描述。如需下载不同于 `v0.20.1-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:v0.20.1` 来下载 RAGFlow 镜像的 `v0.20.1` 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
@@ -196,8 +196,8 @@
 
    | RAGFlow image tag | Image size (GB) | Has embedding models? | Stable?                  |
    | ----------------- | --------------- | --------------------- | ------------------------ |
-   | v0.20.1           | &approx;9       | :heavy_check_mark:    | Stable release           |
-   | v0.20.1-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 |
 
@@ -301,7 +301,7 @@ docker build --platform linux/amd64 --build-arg NEED_MIRROR=1 -f Dockerfile -t i
 
 ## 🔨 以源代码启动服务
 
-1. 安装 uv。如已经安装，可跳过本步骤：
+1. 安装 `uv` 和 `pre-commit`。如已经安装，可跳过本步骤：
 
    ```bash
    pipx install uv pre-commit

agent/canvas.py

@@ -29,83 +29,52 @@ from api.utils import get_uuid, hash_str2int
 from rag.prompts.prompts import chunks_format
 from rag.utils.redis_conn import REDIS_CONN
 
-
-class Canvas:
+class Graph:
     """
-    dsl = {
-        "components": {
-            "begin": {
-                "obj":{
-                    "component_name": "Begin",
-                    "params": {},
-                },
-                "downstream": ["answer_0"],
-                "upstream": [],
-            },
-            "retrieval_0": {
-                "obj": {
-                    "component_name": "Retrieval",
-                    "params": {}
-                },
-                "downstream": ["generate_0"],
-                "upstream": ["answer_0"],
-            },
-            "generate_0": {
-                "obj": {
-                    "component_name": "Generate",
-                    "params": {}
-                },
-                "downstream": ["answer_0"],
-                "upstream": ["retrieval_0"],
-            }
-        },
-        "history": [],
-        "path": ["begin"],
-        "retrieval": {"chunks": [], "doc_aggs": []},
-        "globals": {
-            "sys.query": "",
-            "sys.user_id": tenant_id,
-            "sys.conversation_turns": 0,
-            "sys.files": []
-        }
-    }
-    """
-
-    def __init__(self, dsl: str, tenant_id=None, task_id=None):
-        self.path = []
-        self.history = []
-        self.components = {}
-        self.error = ""
-        self.globals = {
-            "sys.query": "",
-            "sys.user_id": tenant_id,
-            "sys.conversation_turns": 0,
-            "sys.files": []
-        }
-        self.dsl = json.loads(dsl) if dsl else {
+        dsl = {
             "components": {
                 "begin": {
-                    "obj": {
+                    "obj":{
                         "component_name": "Begin",
-                        "params": {
-                            "prologue": "Hi there!"
-                        }
+                        "params": {},
                     },
-                    "downstream": [],
+                    "downstream": ["answer_0"],
                     "upstream": [],
-                    "parent_id": ""
+                },
+                "retrieval_0": {
+                    "obj": {
+                        "component_name": "Retrieval",
+                        "params": {}
+                    },
+                    "downstream": ["generate_0"],
+                    "upstream": ["answer_0"],
+                },
+                "generate_0": {
+                    "obj": {
+                        "component_name": "Generate",
+                        "params": {}
+                    },
+                    "downstream": ["answer_0"],
+                    "upstream": ["retrieval_0"],
                 }
             },
             "history": [],
-            "path": [],
-            "retrieval": [],
+            "path": ["begin"],
+            "retrieval": {"chunks": [], "doc_aggs": []},
             "globals": {
                 "sys.query": "",
-                "sys.user_id": "",
+                "sys.user_id": tenant_id,
                 "sys.conversation_turns": 0,
                 "sys.files": []
             }
         }
+        """
+
+    def __init__(self, dsl: str, tenant_id=None, task_id=None):
+        self.path = []
+        self.components = {}
+        self.error = ""
+        self.dsl = json.loads(dsl)
         self._tenant_id = tenant_id
         self.task_id = task_id if task_id else get_uuid()
         self.load()
@@ -116,8 +85,6 @@ class Canvas:
         for k, cpn in self.components.items():
             cpn_nms.add(cpn["obj"]["component_name"])
 
-        assert "Begin" in cpn_nms, "There have to be an 'Begin' component."
-
         for k, cpn in self.components.items():
             cpn_nms.add(cpn["obj"]["component_name"])
             param = component_class(cpn["obj"]["component_name"] + "Param")()
@@ -130,18 +97,10 @@ class Canvas:
             cpn["obj"] = component_class(cpn["obj"]["component_name"])(self, k, param)
 
         self.path = self.dsl["path"]
-        self.history = self.dsl["history"]
-        self.globals = self.dsl["globals"]
-        self.retrieval = self.dsl["retrieval"]
-        self.memory = self.dsl.get("memory", [])
 
     def __str__(self):
         self.dsl["path"] = self.path
-        self.dsl["history"] = self.history
-        self.dsl["globals"] = self.globals
         self.dsl["task_id"] = self.task_id
-        self.dsl["retrieval"] = self.retrieval
-        self.dsl["memory"] = self.memory
         dsl = {
             "components": {}
         }
@@ -160,14 +119,79 @@ class Canvas:
                 dsl["components"][k][c] = deepcopy(cpn[c])
         return json.dumps(dsl, ensure_ascii=False)
 
-    def reset(self, mem=False):
+    def reset(self):
         self.path = []
+        for k, cpn in self.components.items():
+            self.components[k]["obj"].reset()
+        try:
+            REDIS_CONN.delete(f"{self.task_id}-logs")
+        except Exception as e:
+            logging.exception(e)
+
+    def get_component_name(self, cid):
+        for n in self.dsl.get("graph", {}).get("nodes", []):
+            if cid == n["id"]:
+                return n["data"]["name"]
+        return ""
+
+    def run(self, **kwargs):
+        raise NotImplementedError()
+
+    def get_component(self, cpn_id) -> Union[None, dict[str, Any]]:
+        return self.components.get(cpn_id)
+
+    def get_component_obj(self, cpn_id) -> ComponentBase:
+        return self.components.get(cpn_id)["obj"]
+
+    def get_component_type(self, cpn_id) -> str:
+        return self.components.get(cpn_id)["obj"].component_name
+
+    def get_component_input_form(self, cpn_id) -> dict:
+        return self.components.get(cpn_id)["obj"].get_input_form()
+
+    def get_tenant_id(self):
+        return self._tenant_id
+
+
+class Canvas(Graph):
+
+    def __init__(self, dsl: str, tenant_id=None, task_id=None):
+        self.globals = {
+            "sys.query": "",
+            "sys.user_id": tenant_id,
+            "sys.conversation_turns": 0,
+            "sys.files": []
+        }
+        super().__init__(dsl, tenant_id, task_id)
+
+    def load(self):
+        super().load()
+        self.history = self.dsl["history"]
+        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", [])
+
+    def __str__(self):
+        self.dsl["history"] = self.history
+        self.dsl["retrieval"] = self.retrieval
+        self.dsl["memory"] = self.memory
+        return super().__str__()
+
+    def reset(self, mem=False):
+        super().reset()
         if not mem:
             self.history = []
             self.retrieval = []
             self.memory = []
-        for k, cpn in self.components.items():
-            self.components[k]["obj"].reset()
 
         for k in self.globals.keys():
             if isinstance(self.globals[k], str):
@@ -183,22 +207,13 @@ class Canvas:
             else:
                 self.globals[k] = None
 
-        try:
-            REDIS_CONN.delete(f"{self.task_id}-logs")
-        except Exception as e:
-            logging.exception(e)
-
-    def get_component_name(self, cid):
-        for n in self.dsl.get("graph", {}).get("nodes", []):
-            if cid == n["id"]:
-                return n["data"]["name"]
-        return ""
-
     def run(self, **kwargs):
         st = time.perf_counter()
         self.message_id = get_uuid()
         created_at = int(time.time())
         self.add_user_input(kwargs.get("query"))
+        for k, cpn in self.components.items():
+            self.components[k]["obj"].reset(True)
 
         for k in kwargs.keys():
             if k in ["query", "user_id", "files"] and kwargs[k]:
@@ -377,18 +392,6 @@ class Canvas:
                        })
             self.history.append(("assistant", self.get_component_obj(self.path[-1]).output()))
 
-    def get_component(self, cpn_id) -> Union[None, dict[str, Any]]:
-        return self.components.get(cpn_id)
-
-    def get_component_obj(self, cpn_id) -> ComponentBase:
-        return self.components.get(cpn_id)["obj"]
-
-    def get_component_type(self, cpn_id) -> str:
-        return self.components.get(cpn_id)["obj"].component_name
-
-    def get_component_input_form(self, cpn_id) -> dict:
-        return self.components.get(cpn_id)["obj"].get_input_form()
-
     def is_reff(self, exp: str) -> bool:
         exp = exp.strip("{").strip("}")
         if exp.find("@") < 0:
@@ -410,14 +413,11 @@ class Canvas:
             raise Exception(f"Can't find variable: '{cpn_id}@{var_nm}'")
         return cpn["obj"].output(var_nm)
 
-    def get_tenant_id(self):
-        return self._tenant_id
-
     def get_history(self, window_size):
         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:
@@ -427,39 +427,12 @@ class Canvas:
     def add_user_input(self, question):
         self.history.append(("user", question))
 
-    def _find_loop(self, max_loops=6):
-        path = self.path[-1][::-1]
-        if len(path) < 2:
-            return False
-
-        for i in range(len(path)):
-            if path[i].lower().find("answer") == 0 or path[i].lower().find("iterationitem") == 0:
-                path = path[:i]
-                break
-
-        if len(path) < 2:
-            return False
-
-        for loc in range(2, len(path) // 2):
-            pat = ",".join(path[0:loc])
-            path_str = ",".join(path)
-            if len(pat) >= len(path_str):
-                return False
-            loop = max_loops
-            while path_str.find(pat) == 0 and loop >= 0:
-                loop -= 1
-                if len(pat)+1 >= len(path_str):
-                    return False
-                path_str = path_str[len(pat)+1:]
-            if loop < 0:
-                pat = " => ".join([p.split(":")[0] for p in path[0:loc]])
-                return pat + " => " + pat
-
-        return False
-
     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 +457,7 @@ 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 +466,16 @@ 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/__init__.py

@@ -50,8 +50,9 @@ del _package_path, _import_submodules, _extract_classes_from_module
 
 
 def component_class(class_name):
-    m = importlib.import_module("agent.component")
-    try:
-        return getattr(m, class_name)
-    except Exception:
-        return getattr(importlib.import_module("agent.tools"), class_name)
+    for mdl in ["agent.component", "agent.tools", "rag.flow"]:
+        try:
+            return getattr(importlib.import_module(mdl), class_name)
+        except Exception:
+            pass
+    assert False, f"Can't import {class_name}"

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

@@ -16,7 +16,7 @@
 
 import re
 import time
-from abc import ABC, abstractmethod
+from abc import ABC
 import builtins
 import json
 import os
@@ -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 = ""
@@ -410,8 +410,8 @@ class ComponentBase(ABC):
         )
 
     def __init__(self, canvas, id, param: ComponentParamBase):
-        from agent.canvas import Canvas  # Local import to avoid cyclic dependency
-        assert isinstance(canvas, Canvas), "canvas must be an instance of Canvas"
+        from agent.canvas import Graph  # Local import to avoid cyclic dependency
+        assert isinstance(canvas, Graph), "canvas must be an instance of Canvas"
         self._canvas = canvas
         self._id = id
         self._param = param
@@ -448,9 +448,11 @@ class ComponentBase(ABC):
     def error(self):
         return self._param.outputs.get("_ERROR", {}).get("value")
 
-    def reset(self):
+    def reset(self, only_output=False):
         for k in self._param.outputs.keys():
             self._param.outputs[k]["value"] = None
+        if only_output:
+            return
         for k in self._param.inputs.keys():
             self._param.inputs[k]["value"] = None
         self._param.debug_inputs = {}
@@ -479,7 +481,7 @@ 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] = {
@@ -526,11 +528,19 @@ class ComponentBase(ABC):
         cpn_nms = self._canvas.get_component(self._id)['upstream']
         return cpn_nms
 
+    def get_downstream(self) -> List[str]:
+        cpn_nms = self._canvas.get_component(self._id)['downstream']
+        return cpn_nms
+
     @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
 
@@ -550,6 +560,5 @@ class ComponentBase(ABC):
     def set_exception_default_value(self):
         self.set_output("result", self.get_exception_default_value())
 
-    @abstractmethod
     def thoughts(self) -> str:
-        ...
+        raise NotImplementedError()

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/choose_your_knowledge_base_agent.json

@@ -1,8 +1,12 @@
 {
     "id": 19,
-    "title": "Choose Your Knowledge Base Agent",
-    "description": "Select your desired knowledge base from the dropdown menu. The Agent will only retrieve from the selected knowledge base and use this content  to generate responses.",
-    "canvas_type": "Agent", 
+    "title": {
+		"en": "Choose Your Knowledge Base Agent",
+		"zh": "选择知识库智能体"},
+    "description": {
+		"en": "Select your desired knowledge base from the dropdown menu. The Agent will only retrieve from the selected knowledge base and use this content  to generate responses.",
+		"zh": "从下拉菜单中选择知识库，智能体将仅根据所选知识库内容生成回答。"},
+	"canvas_type": "Agent", 
     "dsl": {
 		"components": {
 			"Agent:BraveParksJoke": {

agent/templates/choose_your_knowledge_base_workflow.json

@@ -1,8 +1,12 @@
 {
     "id": 18,
-    "title": "Choose Your Knowledge Base Workflow",
-    "description": "Select your desired knowledge base from the dropdown menu. The retrieval assistant will only use data from your selected knowledge base to generate responses.",
-    "canvas_type": "Other",
+    "title": {
+		"en": "Choose Your Knowledge Base Workflow",
+		"zh": "选择知识库工作流"},
+    "description": {
+		"en": "Select your desired knowledge base from the dropdown menu. The retrieval assistant will only use data from your selected knowledge base to generate responses.",
+		"zh": "从下拉菜单中选择知识库，工作流将仅根据所选知识库内容生成回答。"},
+	"canvas_type": "Other",
     "dsl": {
 		"components": {
 			"Agent:ProudDingosShout": {

agent/templates/customer_review_analysis.json

@@ -1,9 +1,13 @@
 
 {
     "id": 11,
-    "title": "Customer Review Analysis",
-    "description": "Automatically classify customer reviews using LLM (Large Language Model) and route them via email to the relevant departments.",
-    "canvas_type": "Customer Support",
+    "title": {
+		"en": "Customer Review Analysis",
+		"zh": "客户评价分析"},
+    "description": {
+		"en": "Automatically classify customer reviews using LLM (Large Language Model) and route them via email to the relevant departments.",
+		"zh": "大模型将自动分类客户评价，并通过电子邮件将结果发送到相关部门。"},
+	"canvas_type": "Customer Support",
     "dsl": {
 		"components": {
 			"Categorize:FourTeamsFold": {

agent/templates/customer_support.json

@@ -1,8 +1,12 @@
 
 {
     "id": 10,
-    "title": "Customer Support",
-    "description": "This is an intelligent customer service processing system workflow based on user intent classification. It uses LLM to identify user demand types and transfers them to the corresponding professional agent for processing.",
+    "title": {
+        "en":"Customer Support",
+        "zh": "客户支持"},
+    "description": {
+        "en": "This is an intelligent customer service processing system workflow based on user intent classification. It uses LLM to identify user demand types and transfers them to the corresponding professional agent for processing.",
+        "zh": "工作流系统，用于智能客服场景。基于用户意图分类。使用大模型识别用户需求类型，并将需求转移给相应的智能体进行处理。"},
     "canvas_type": "Customer Support",
     "dsl": {
             "components": {

agent/templates/cv_analysis_and_candidate_evaluation.json

@@ -1,8 +1,12 @@
 
 {
     "id": 15,
-    "title": "CV Analysis and Candidate Evaluation",
-    "description": "This is a workflow that helps companies evaluate resumes, HR uploads a job description first, then submits multiple resumes via the chat window for evaluation.",
+    "title": {
+        "en": "CV Analysis and Candidate Evaluation",
+        "zh": "简历分析和候选人评估"},
+    "description": {
+        "en": "This is a workflow that helps companies evaluate resumes, HR uploads a job description first, then submits multiple resumes via the chat window for evaluation.",
+        "zh": "帮助公司评估简历的工作流。HR首先上传职位描述，通过聊天窗口提交多份简历进行评估。"},
     "canvas_type": "Other",
     "dsl": {
             "components": {

agent/templates/deep_research.json

@@ -1,8 +1,12 @@
       
 {
     "id": 1,
-    "title": "Deep Research",
-    "description": "For professionals in sales, marketing, policy, or consulting, the Multi-Agent Deep Research Agent conducts structured, multi-step investigations across diverse sources and delivers consulting-style reports with clear citations.",
+    "title": {
+        "en": "Deep Research",
+        "zh": "深度研究"},
+    "description": {
+        "en": "For professionals in sales, marketing, policy, or consulting, the Multi-Agent Deep Research Agent conducts structured, multi-step investigations across diverse sources and delivers consulting-style reports with clear citations.",
+        "zh": "专为销售、市场、政策或咨询领域的专业人士设计，多智能体的深度研究会结合多源信息进行结构化、多步骤地回答问题，并附带有清晰的引用。"},
     "canvas_type": "Recommended",
     "dsl": {
             "components": {

agent/templates/deep_search_r.json

@@ -1,8 +1,12 @@
 
 {
     "id": 6,
-    "title": "Deep Research",
-    "description": "For professionals in sales, marketing, policy, or consulting, the Multi-Agent Deep Research Agent conducts structured, multi-step investigations across diverse sources and delivers consulting-style reports with clear citations.",
+    "title": {
+        "en": "Deep Research",
+        "zh": "深度研究"},
+    "description": {
+        "en": "For professionals in sales, marketing, policy, or consulting, the Multi-Agent Deep Research Agent conducts structured, multi-step investigations across diverse sources and delivers consulting-style reports with clear citations.",
+        "zh": "专为销售、市场、政策或咨询领域的专业人士设计，多智能体的深度研究会结合多源信息进行结构化、多步骤地回答问题，并附带有清晰的引用。"},
     "canvas_type": "Agent",
     "dsl": {
             "components": {

agent/templates/generate_SEO_blog.json

@@ -1,7 +1,11 @@
 {
     "id": 8,
-    "title": "Generate SEO Blog",
-    "description": "This is a multi-agent version of the SEO blog generation workflow. It simulates a small team of AI “writers”, where each agent plays a specialized role — just like a real editorial team.",
+    "title": {
+        "en": "Generate SEO Blog",
+        "zh": "生成SEO博客"},
+    "description": {
+        "en": "This is a multi-agent version of the SEO blog generation workflow. It simulates a small team of AI “writers”, where each agent plays a specialized role — just like a real editorial team.",
+        "zh": "多智能体架构可根据简单的用户输入自动生成完整的SEO博客文章。模拟小型“作家”团队，其中每个智能体扮演一个专业角色——就像真正的编辑团队。"},
     "canvas_type": "Agent",
     "dsl": {
             "components": {

agent/templates/image_lingo.json

@@ -1,7 +1,11 @@
 {
     "id": 13,
-    "title": "ImageLingo",
-    "description": "ImageLingo lets you snap any photo containing text—menus, signs, or documents—and instantly recognize and translate it into your language of choice using advanced AI-powered translation technology.",
+    "title": {
+        "en": "ImageLingo",
+        "zh": "图片解析"},
+    "description": {
+        "en": "ImageLingo lets you snap any photo containing text—menus, signs, or documents—and instantly recognize and translate it into your language of choice using advanced AI-powered translation technology.",
+        "zh": "多模态大模型允许您拍摄任何包含文本的照片——菜单、标志或文档——立即识别并转换成您选择的语言。"},
     "canvas_type": "Consumer App",
     "dsl": {
             "components": {

agent/templates/knowledge_base_report.json

@@ -0,0 +1,331 @@
+{
+    "id": 20,
+    "title": {
+        "en": "Report Agent Using Knowledge Base",
+        "zh": "知识库检索智能体"},
+    "description": {
+        "en": "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",
+        "zh": "一个使用本地知识库的报告生成助手，具备高级能力，包括任务规划、推理和反思性分析。推荐用于学术研究论文问答。"},
+    "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/knowledge_base_report_r.json

@@ -0,0 +1,331 @@
+{
+    "id": 21,
+    "title": {
+        "en": "Report Agent Using Knowledge Base", 
+        "zh": "知识库检索智能体"},
+    "description": {
+        "en": "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",
+        "zh": "一个使用本地知识库的报告生成助手，具备高级能力，包括任务规划、推理和反思性分析。推荐用于学术研究论文问答。"},
+    "canvas_type": "Recommended",
+    "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

@@ -1,7 +1,11 @@
 {
     "id": 12,
-    "title": "Generate SEO Blog",
-    "description": "This workflow automatically generates a complete SEO-optimized blog article based on a simple user input. You don’t need any writing experience. Just provide a topic or short request — the system will handle the rest.",
+    "title": {
+        "en": "Generate SEO Blog",
+        "zh": "生成SEO博客"},
+    "description": {
+        "en": "This workflow automatically generates a complete SEO-optimized blog article based on a simple user input. You don’t need any writing experience. Just provide a topic or short request — the system will handle the rest.",
+        "zh": "此工作流根据简单的用户输入自动生成完整的SEO博客文章。你无需任何写作经验，只需提供一个主题或简短请求，系统将处理其余部分。"},
     "canvas_type": "Marketing",
     "dsl": {
             "components": {

agent/templates/seo_blog.json

@@ -1,7 +1,11 @@
 {
     "id": 4,
-    "title": "Generate SEO Blog",
-    "description": "This workflow automatically generates a complete SEO-optimized blog article based on a simple user input. You don’t need any writing experience. Just provide a topic or short request — the system will handle the rest.",
+    "title": {
+        "en": "Generate SEO Blog",
+        "zh": "生成SEO博客"},
+    "description": {
+        "en": "This workflow automatically generates a complete SEO-optimized blog article based on a simple user input. You don’t need any writing experience. Just provide a topic or short request — the system will handle the rest.",
+        "zh": "此工作流根据简单的用户输入自动生成完整的SEO博客文章。你无需任何写作经验，只需提供一个主题或简短请求，系统将处理其余部分。"},
     "canvas_type": "Recommended",
     "dsl": {
             "components": {

agent/templates/sql_assistant.json

@@ -1,7 +1,11 @@
 {
     "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. ",
+    "title": {
+        "en": "SQL Assistant",
+        "zh": "SQL助理"},
+    "description": {
+        "en": "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. ",
+        "zh": "用户能够将简单文本问题转化为完整的SQL查询并输出结果。只需输入您的问题（例如，“展示上个季度前十名按收入排序的产品”），SQL助理就会生成精确的SQL语句，对其运行您的数据库，并几秒钟内返回结果。"},
     "canvas_type": "Marketing",
     "dsl": {
             "components": {
@@ -206,7 +210,7 @@
                             "enablePrologue": true,
                             "inputs": {},
                             "mode": "conversational",
-                            "prologue": "Hi! I'm your SQL assistant, what can I do for you?"
+                            "prologue": "Hi! I'm your SQL assistant. What can I do for you?"
                         }
                     },
                     "upstream": []
@@ -319,7 +323,7 @@
                                 "enablePrologue": true,
                                 "inputs": {},
                                 "mode": "conversational",
-                                "prologue": "Hi! I'm your SQL assistant, what can I do for you?"
+                                "prologue": "Hi! I'm your SQL assistant. What can I do for you?"
                             },
                             "label": "Begin",
                             "name": "begin"

agent/templates/technical_docs_qa.json

@@ -1,8 +1,12 @@
 
 {
     "id": 9,
-    "title": "Technical Docs QA",
-    "description": "This is a document question-and-answer system based on a knowledge base. When a user asks a question, it retrieves relevant document content to provide accurate answers.",
+    "title": {
+        "en": "Technical Docs QA",
+        "zh": "技术文档问答"},
+    "description": {
+        "en": "This is a document question-and-answer system based on a knowledge base. When a user asks a question, it retrieves relevant document content to provide accurate answers.",
+        "zh": "基于知识库的文档问答系统，当用户提出问题时，会检索相关本地文档并提供准确回答。"},
     "canvas_type": "Customer Support",
     "dsl": {
             "components": {

agent/templates/trip_planner.json

@@ -1,9 +1,13 @@
 
 {
     "id": 14,
-    "title": "Trip Planner",
-    "description": "This smart trip planner utilizes LLM technology to automatically generate customized travel itineraries, with optional tool integration for enhanced reliability.",
-    "canvas_type": "Consumer App",
+    "title": {
+		"en": "Trip Planner",
+		"zh": "旅行规划"},
+    "description": {
+		"en": "This smart trip planner utilizes LLM technology to automatically generate customized travel itineraries, with optional tool integration for enhanced reliability.",
+		"zh": "智能旅行规划将利用大模型自动生成定制化的旅行行程，附带可选工具集成，以增强可靠性。"},
+	"canvas_type": "Consumer App",
     "dsl": {
 		"components": {
 			"Agent:OddGuestsPump": {

agent/templates/web_search_assistant.json

@@ -1,9 +1,13 @@
 
 {
     "id": 16,
-    "title": "WebSearch Assistant",
-    "description": "A chat assistant template that integrates information extracted from a knowledge base and web searches to respond to queries. Let's start by setting up your knowledge base in 'Retrieval'!",
-    "canvas_type": "Other",
+    "title": {
+		"en": "WebSearch Assistant",
+		"zh": "网页搜索助手"},
+    "description": {
+		"en": "A chat assistant template that integrates information extracted from a knowledge base and web searches to respond to queries. Let's start by setting up your knowledge base in 'Retrieval'!",
+		"zh": "集成了从知识库和网络搜索中提取的信息回答用户问题。让我们从设置您的知识库开始检索！"},
+	"canvas_type": "Other",
     "dsl": {
 		"components": {
 			"Agent:SmartSchoolsCross": {

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/crawler.py

@@ -16,9 +16,8 @@
 from abc import ABC
 import asyncio
 from crawl4ai import AsyncWebCrawler
-
 from agent.tools.base import ToolParamBase, ToolBase
-from api.utils.web_utils import is_valid_url
+
 
 
 class CrawlerParam(ToolParamBase):
@@ -39,6 +38,7 @@ class Crawler(ToolBase, ABC):
     component_name = "Crawler"
 
     def _run(self, history, **kwargs):
+        from api.utils.web_utils import is_valid_url
         ans = self.get_input()
         ans = " - ".join(ans["content"]) if "content" in ans else ""
         if not is_valid_url(ans):

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, self._canvas._tenant_id)
-            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)
 

agent/tools/searxng.py

@@ -0,0 +1,156 @@
+#
+#  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
+import os
+import time
+from abc import ABC
+import requests
+from agent.tools.base import ToolMeta, ToolParamBase, ToolBase
+from api.utils.api_utils import timeout
+
+
+class SearXNGParam(ToolParamBase):
+    """
+    Define the SearXNG component parameters.
+    """
+
+    def __init__(self):
+        self.meta: ToolMeta = {
+            "name": "searxng_search",
+            "description": "SearXNG is a privacy-focused metasearch engine that aggregates results from multiple search engines without tracking users. It provides comprehensive web search capabilities.",
+            "parameters": {
+                "query": {
+                    "type": "string",
+                    "description": "The search keywords to execute with SearXNG. The keywords should be the most important words/terms(includes synonyms) from the original request.",
+                    "default": "{sys.query}",
+                    "required": True
+                },
+                "searxng_url": {
+                    "type": "string",
+                    "description": "The base URL of your SearXNG instance (e.g., http://localhost:4000). This is required to connect to your SearXNG server.",
+                    "required": False,
+                    "default": ""
+                }
+            }
+        }
+        super().__init__()
+        self.top_n = 10
+        self.searxng_url = ""
+
+    def check(self):
+        # Keep validation lenient so opening try-run panel won't fail without URL.
+        # Coerce top_n to int if it comes as string from UI.
+        try:
+            if isinstance(self.top_n, str):
+                self.top_n = int(self.top_n.strip())
+        except Exception:
+            pass
+        self.check_positive_integer(self.top_n, "Top N")
+
+    def get_input_form(self) -> dict[str, dict]:
+        return {
+            "query": {
+                "name": "Query",
+                "type": "line"
+            },
+            "searxng_url": {
+                "name": "SearXNG URL",
+                "type": "line",
+                "placeholder": "http://localhost:4000"
+            }
+        }
+
+
+class SearXNG(ToolBase, ABC):
+    component_name = "SearXNG"
+
+    @timeout(os.environ.get("COMPONENT_EXEC_TIMEOUT", 12))
+    def _invoke(self, **kwargs):
+        # Gracefully handle try-run without inputs
+        query = kwargs.get("query")
+        if not query or not isinstance(query, str) or not query.strip():
+            self.set_output("formalized_content", "")
+            return ""
+
+        searxng_url = (kwargs.get("searxng_url") or getattr(self._param, "searxng_url", "") or "").strip()
+        # In try-run, if no URL configured, just return empty instead of raising
+        if not searxng_url:
+            self.set_output("formalized_content", "")
+            return ""
+
+        last_e = ""
+        for _ in range(self._param.max_retries+1):
+            try:
+                # 构建搜索参数
+                search_params = {
+                    'q': query,
+                    'format': 'json',
+                    'categories': 'general',
+                    'language': 'auto',
+                    'safesearch': 1,
+                    'pageno': 1
+                }
+
+                # 发送搜索请求
+                response = requests.get(
+                    f"{searxng_url}/search",
+                    params=search_params,
+                    timeout=10
+                )
+                response.raise_for_status()
+                
+                data = response.json()
+                
+                # 验证响应数据
+                if not data or not isinstance(data, dict):
+                    raise ValueError("Invalid response from SearXNG")
+                
+                results = data.get("results", [])
+                if not isinstance(results, list):
+                    raise ValueError("Invalid results format from SearXNG")
+                
+                # 限制结果数量
+                results = results[:self._param.top_n]
+                
+                # 处理搜索结果
+                self._retrieve_chunks(results,
+                                      get_title=lambda r: r.get("title", ""),
+                                      get_url=lambda r: r.get("url", ""),
+                                      get_content=lambda r: r.get("content", ""))
+                
+                self.set_output("json", results)
+                return self.output("formalized_content")
+
+            except requests.RequestException as e:
+                last_e = f"Network error: {e}"
+                logging.exception(f"SearXNG network error: {e}")
+                time.sleep(self._param.delay_after_error)
+            except Exception as e:
+                last_e = str(e)
+                logging.exception(f"SearXNG error: {e}")
+                time.sleep(self._param.delay_after_error)
+
+        if last_e:
+            self.set_output("_ERROR", last_e)
+            return f"SearXNG error: {last_e}"
+
+        assert False, self.output()
+
+    def thoughts(self) -> str:
+        return """
+Keywords: {} 
+Searching with SearXNG for relevant results...
+                """.format(self.get_input().get("query", "-_-!"))

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

@@ -74,11 +74,11 @@ 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()
@@ -90,7 +90,7 @@ def save():
                 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)
@@ -115,6 +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.")
@@ -347,7 +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)
@@ -369,7 +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())
@@ -379,7 +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))
@@ -387,10 +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)
 

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
 
@@ -90,6 +93,7 @@ def list_chunk():
 def get():
     chunk_id = request.args["chunk_id"]
     try:
+        chunk = None
         tenants = UserTenantService.query(user_id=current_user.id)
         if not tenants:
             return get_data_error_result(message="Tenant not found!")
@@ -288,13 +292,26 @@ 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 +344,9 @@ 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
 
 
@@ -66,8 +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,
-                "reference":[{}],}
+        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:
@@ -174,6 +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:
@@ -187,23 +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"
@@ -221,7 +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:
@@ -317,10 +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"
@@ -339,18 +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)
@@ -361,41 +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,
         [
@@ -407,6 +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
@@ -41,6 +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)
@@ -51,11 +61,12 @@ 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 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.")
+            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"]:
@@ -85,6 +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,

api/apps/document_app.py

@@ -681,6 +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

@@ -351,6 +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):
@@ -364,3 +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

@@ -17,7 +17,8 @@ import logging
 import json
 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
@@ -57,6 +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:
@@ -73,7 +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})
@@ -204,6 +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](
@@ -221,7 +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!"}], {
@@ -239,7 +243,7 @@ def add_llm():
                 model_name=mdl_nm,
                 base_url=llm["api_base"]
             )
-            arr, tc = mdl.similarity("Hello~ Ragflower!", ["Hi, there!", "Ohh, my friend!"])
+            arr, tc = mdl.similarity("Hello~ RAGFlower!", ["Hi, there!", "Ohh, my friend!"])
             if len(arr) == 0:
                 raise Exception("Not known.")
         except KeyError:
@@ -267,7 +271,7 @@ def add_llm():
             key=llm["api_key"], model_name=mdl_nm, base_url=llm["api_base"]
         )
         try:
-            for resp in mdl.tts("Hello~ Ragflower!"):
+            for resp in mdl.tts("Hello~ RAGFlower!"):
                 pass
         except RuntimeError as e:
             msg += f"\nFail to access model({factory}/{mdl_nm})." + str(e)
@@ -312,12 +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
@@ -325,13 +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"],
@@ -352,7 +356,7 @@ 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)

api/apps/mcp_server_app.py

@@ -82,7 +82,7 @@ def create() -> Response:
 
     server_name = req.get("name", "")
     if not server_name or len(server_name.encode("utf-8")) > 255:
-        return get_data_error_result(message=f"Invaild MCP name or length is {len(server_name)} which is large than 255.")
+        return get_data_error_result(message=f"Invalid MCP name or length is {len(server_name)} which is large than 255.")
 
     e, _ = MCPServerService.get_by_name_and_tenant(name=server_name, tenant_id=current_user.id)
     if e:
@@ -90,7 +90,7 @@ def create() -> Response:
 
     url = req.get("url", "")
     if not url:
-        return get_data_error_result(message="Invaild url.")
+        return get_data_error_result(message="Invalid url.")
 
     headers = safe_json_parse(req.get("headers", {}))
     req["headers"] = headers
@@ -141,10 +141,10 @@ def update() -> Response:
         return get_data_error_result(message="Unsupported MCP server type.")
     server_name = req.get("name", mcp_server.name)
     if server_name and len(server_name.encode("utf-8")) > 255:
-        return get_data_error_result(message=f"Invaild MCP name or length is {len(server_name)} which is large than 255.")
+        return get_data_error_result(message=f"Invalid MCP name or length is {len(server_name)} which is large than 255.")
     url = req.get("url", mcp_server.url)
     if not url:
-        return get_data_error_result(message="Invaild url.")
+        return get_data_error_result(message="Invalid url.")
 
     headers = safe_json_parse(req.get("headers", mcp_server.headers))
     req["headers"] = headers
@@ -218,7 +218,7 @@ def import_multiple() -> Response:
                 continue
 
             if not server_name or len(server_name.encode("utf-8")) > 255:
-                results.append({"server": server_name, "success": False, "message": f"Invaild MCP name or length is {len(server_name)} which is large than 255."})
+                results.append({"server": server_name, "success": False, "message": f"Invalid MCP name or length is {len(server_name)} which is large than 255."})
                 continue
 
             base_name = server_name
@@ -409,7 +409,7 @@ def test_mcp() -> Response:
 
     url = req.get("url", "")
     if not url:
-        return get_data_error_result(message="Invaild MCP url.")
+        return get_data_error_result(message="Invalid MCP url.")
 
     server_type = req.get("server_type", "")
     if server_type not in VALID_MCP_SERVER_TYPES:

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

@@ -24,6 +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
@@ -37,18 +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,
@@ -59,6 +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])
         )
 
@@ -93,3 +100,20 @@ def retrieval(tenant_id):
             )
         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

@@ -32,7 +32,8 @@ 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

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
@@ -69,11 +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.")
@@ -82,45 +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)
@@ -445,7 +414,7 @@ def agents_completion_openai_compatibility(tenant_id, agent_id):
                 tenant_id,
                 agent_id,
                 question,
-                session_id=req.get("id", req.get("metadata", {}).get("id", "")),
+                session_id=req.get("session_id", req.get("id", "") or req.get("metadata", {}).get("id", "")),
                 stream=True,
                 **req,
             ),
@@ -463,7 +432,7 @@ def agents_completion_openai_compatibility(tenant_id, agent_id):
                 tenant_id,
                 agent_id,
                 question,
-                session_id=req.get("id", req.get("metadata", {}).get("id", "")),
+                session_id=req.get("session_id", req.get("id", "") or req.get("metadata", {}).get("id", "")),
                 stream=False,
                 **req,
             )
@@ -476,7 +445,6 @@ def agents_completion_openai_compatibility(tenant_id, agent_id):
 def agent_completions(tenant_id, agent_id):
     req = request.json
 
-    ans = {}
     if req.get("stream", True):
 
         def generate():
@@ -487,7 +455,7 @@ def agent_completions(tenant_id, agent_id):
                     except Exception:
                         continue
 
-                if ans.get("event") != "message":
+                if ans.get("event") not in ["message", "message_end"]:
                     continue
 
                 yield answer
@@ -501,12 +469,25 @@ def agent_completions(tenant_id, agent_id):
         resp.headers.add_header("Content-Type", "text/event-stream; charset=utf-8")
         return resp
 
+    full_content = ""
+    reference = {}
+    final_ans = ""
     for answer in agent_completion(tenant_id=tenant_id, agent_id=agent_id, **req):
         try:
-            ans = json.loads(answer[5:])  # remove "data:"
+            ans = json.loads(answer[5:])
+
+            if ans["event"] == "message":
+                full_content += ans["data"]["content"]
+
+            if ans.get("data", {}).get("reference", None):
+                reference.update(ans["data"]["reference"])
+
+            final_ans = ans
         except Exception as e:
             return get_result(data=f"**ERROR**: {str(e)}")
-    return get_result(data=ans)
+    final_ans["data"]["content"] = full_content
+    final_ans["data"]["reference"] = reference
+    return get_result(data=final_ans)
 
 
 @manager.route("/chats/<chat_id>/sessions", methods=["GET"])  # noqa: F821
@@ -589,16 +570,24 @@ 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")),
@@ -832,6 +821,29 @@ 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
@@ -871,11 +883,231 @@ 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"),
-            "prologue": canvas.get_prologue()
-        }
+    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
@@ -44,10 +43,10 @@ def create():
         return get_data_error_result(message=f"Search name length is {len(search_name)} which is large than 255.")
     e, _ = TenantService.get_by_id(current_user.id)
     if not e:
-        return get_data_error_result(message="Authorizationd identity.")
+        return get_data_error_result(message="Authorized 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
@@ -79,7 +78,7 @@ def update():
     tenant_id = req["tenant_id"]
     e, _ = TenantService.get_by_id(tenant_id)
     if not e:
-        return get_data_error_result(message="Authorizationd identity.")
+        return get_data_error_result(message="Authorized identity.")
 
     search_id = req["search_id"]
     if not SearchService.accessible4deletion(search_id, current_user.id):
@@ -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)
@@ -823,9 +824,8 @@ class UserCanvas(DataBaseModel):
 class CanvasTemplate(DataBaseModel):
     id = CharField(max_length=32, primary_key=True)
     avatar = TextField(null=True, help_text="avatar base64 string")
-    title = CharField(max_length=255, null=True, help_text="Canvas title")
-
-    description = TextField(null=True, help_text="Canvas description")
+    title = JSONField(null=True, default=dict, help_text="Canvas title")
+    description = JSONField(null=True, default=dict, help_text="Canvas description")
     canvas_type = CharField(max_length=32, null=True, help_text="Canvas type", index=True)
     dsl = JSONField(null=True, default={})
 
@@ -871,7 +871,7 @@ 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 +880,12 @@ 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 +1016,17 @@ 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
+
+    try:
+        migrate(migrator.alter_column_type("canvas_template", "title", JSONField(null=True, default=dict, help_text="Canvas title")))
+    except Exception:
+        pass
+    try:
+        migrate(migrator.alter_column_type("canvas_template", "description", JSONField(null=True, default=dict, help_text="Canvas description")))
+    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

@@ -134,6 +134,7 @@ class UserCanvasService(CommonService):
             return False
         return True
 
+
 def completion(tenant_id, agent_id, session_id=None, **kwargs):
     query = kwargs.get("query", "") or kwargs.get("question", "")
     files = kwargs.get("files", [])
@@ -163,7 +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)
@@ -211,28 +213,33 @@ def completionOpenAI(tenant_id, agent_id, question, session_id=None, stream=True
                     except Exception as e:
                         logging.exception(f"Agent OpenAI-Compatible completionOpenAI parse answer failed: {e}")
                         continue
-
-                if ans.get("event") != "message":
+                if ans.get("event") not in ["message", "message_end"]:
                     continue
 
-                content_piece = ans["data"]["content"]
+                content_piece = ""
+                if ans["event"] == "message":
+                    content_piece = ans["data"]["content"]
+
                 completion_tokens += len(tiktokenenc.encode(content_piece))
 
-                yield "data: " + json.dumps(
-                    get_data_openai(
+                openai_data = get_data_openai(
                         id=session_id or str(uuid4()),
                         model=agent_id,
                         content=content_piece,
                         prompt_tokens=prompt_tokens,
                         completion_tokens=completion_tokens,
                         stream=True
-                    ),
-                    ensure_ascii=False
-                ) + "\n\n"
+                    )
+
+                if ans.get("data", {}).get("reference", None):
+                    openai_data["choices"][0]["delta"]["reference"] = ans["data"]["reference"]
+
+                yield "data: " + json.dumps(openai_data, ensure_ascii=False) + "\n\n"
 
             yield "data: [DONE]\n\n"
 
         except Exception as e:
+            logging.exception(e)
             yield "data: " + json.dumps(
                 get_data_openai(
                     id=session_id or str(uuid4()),
@@ -250,6 +257,7 @@ def completionOpenAI(tenant_id, agent_id, question, session_id=None, stream=True
     else:
         try:
             all_content = ""
+            reference = {}
             for ans in completion(
                 tenant_id=tenant_id,
                 agent_id=agent_id,
@@ -260,13 +268,18 @@ def completionOpenAI(tenant_id, agent_id, question, session_id=None, stream=True
             ):
                 if isinstance(ans, str):
                     ans = json.loads(ans[5:])
-                if ans.get("event") != "message":
+                if ans.get("event") not in ["message", "message_end"]:
                     continue
-                all_content += ans["data"]["content"]
+
+                if ans["event"] == "message":
+                    all_content += ans["data"]["content"]
+
+                if ans.get("data", {}).get("reference", None):
+                    reference.update(ans["data"]["reference"])
 
             completion_tokens = len(tiktokenenc.encode(all_content))
 
-            yield get_data_openai(
+            openai_data = get_data_openai(
                 id=session_id or str(uuid4()),
                 model=agent_id,
                 prompt_tokens=prompt_tokens,
@@ -276,7 +289,12 @@ def completionOpenAI(tenant_id, agent_id, question, session_id=None, stream=True
                 param=None
             )
 
+            if reference:
+                openai_data["choices"][0]["message"]["reference"] = reference
+
+            yield openai_data
         except Exception as e:
+            logging.exception(e)
             yield get_data_openai(
                 id=session_id or str(uuid4()),
                 model=agent_id,

api/db/services/dialog_service.py

@@ -22,6 +22,7 @@ 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
 
@@ -30,14 +31,18 @@ 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
 
@@ -96,7 +101,6 @@ 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):
@@ -119,6 +123,7 @@ class DialogService(CommonService):
             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"),
@@ -250,6 +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"):
@@ -287,9 +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
@@ -316,6 +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])
 
@@ -323,17 +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)):
@@ -621,7 +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]))
 
@@ -630,30 +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]
@@ -671,7 +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

@@ -13,249 +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]
-
-        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
+    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)
-        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}
+        super().__init__(tenant_id, llm_type, llm_name, lang, **kwargs)
 
     def bind_tools(self, toolcall_session, tools):
         if not self.is_tools:
@@ -323,7 +152,7 @@ class LLMBundle:
 
     def describe_with_prompt(self, image, prompt):
         if self.langfuse:
-            generation = self.language.start_generation(trace_context=self.trace_context, 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):
@@ -376,7 +205,24 @@ 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)
 
+        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.langfuse.start_generation(trace_context=self.trace_context, name="chat", model=self.llm_name, input={"system": system, "history": history})
@@ -384,8 +230,9 @@ class LLMBundle:
         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:
@@ -409,8 +256,8 @@ class LLMBundle:
         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:

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/db/services/user_service.py

@@ -133,6 +133,13 @@ class UserService(CommonService):
                 cls.model.update(user_dict).where(
                     cls.model.id == user_id).execute()
 
+    @classmethod
+    @DB.connection_context()
+    def is_admin(cls, user_id):
+        return cls.model.select().where(
+            cls.model.id == user_id,
+            cls.model.is_superuser == 1).count() > 0
+
 
 class TenantService(CommonService):
     """Service class for managing tenant-related database operations.

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
@@ -74,23 +79,32 @@ 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
 
 
@@ -99,10 +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:
@@ -115,29 +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")
 
@@ -170,12 +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
@@ -210,3 +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}},
@@ -666,7 +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
@@ -681,7 +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:

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

@@ -302,6 +302,20 @@
                     "model_type": "chat",
                     "is_tools": true
                 },
+                {
+                    "llm_name": "qwen-plus-2025-07-28",
+                    "tags": "LLM,CHAT,132k",
+                    "max_tokens": 131072,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "qwen-plus-2025-07-14",
+                    "tags": "LLM,CHAT,132k",
+                    "max_tokens": 131072,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
                 {
                     "llm_name": "qwq-plus-latest",
                     "tags": "LLM,CHAT,132k",
@@ -309,6 +323,20 @@
                     "model_type": "chat",
                     "is_tools": true
                 },
+                {
+                    "llm_name": "qwen-flash",
+                    "tags": "LLM,CHAT,1M",
+                    "max_tokens": 1000000,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
+                {
+                    "llm_name": "qwen-flash-2025-07-28",
+                    "tags": "LLM,CHAT,1M",
+                    "max_tokens": 1000000,
+                    "model_type": "chat",
+                    "is_tools": true
+                },
                 {
                     "llm_name": "qwen3-coder-480b-a35b-instruct",
                     "tags": "LLM,CHAT,256k",
@@ -505,6 +533,24 @@
                     "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"
                 }
             ]
         },
@@ -514,23 +560,65 @@
             "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
@@ -1146,60 +1234,35 @@
                     "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
                 },
                 {

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):
@@ -104,6 +131,12 @@ class RAGFlowExcelParser:
 
         return tb_chunks
 
+    def markdown(self, fnm):
+        import pandas as pd
+        file_like_object = BytesIO(fnm) if not isinstance(fnm, str) else fnm
+        df = pd.read_excel(file_like_object)
+        return df.to_markdown(index=False)
+
     def __call__(self, fnm):
         file_like_object = BytesIO(fnm) if not isinstance(fnm, str) else fnm
         wb = RAGFlowExcelParser._load_excel_to_workbook(file_like_object)

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/pdf_parser.py

@@ -93,6 +93,7 @@ class RAGFlowPdfParser:
                 model_dir, "updown_concat_xgb.model"))
 
         self.page_from = 0
+        self.column_num = 1
 
     def __char_width(self, c):
         return (c["x1"] - c["x0"]) // max(len(c["text"]), 1)
@@ -427,10 +428,18 @@ class RAGFlowPdfParser:
             i += 1
         self.boxes = bxs
 
-    def _naive_vertical_merge(self):
+    def _naive_vertical_merge(self, zoomin=3):
         bxs = Recognizer.sort_Y_firstly(
             self.boxes, np.median(
                 self.mean_height) / 3)
+
+        column_width = np.median([b["x1"] - b["x0"] for b in self.boxes])
+        self.column_num = int(self.page_images[0].size[0] / zoomin / column_width)
+        if column_width < self.page_images[0].size[0] / zoomin / self.column_num:
+            logging.info("Multi-column................... {} {}".format(column_width,
+                  self.page_images[0].size[0] / zoomin / self.column_num))
+            self.boxes = self.sort_X_by_page(self.boxes, column_width / self.column_num)
+
         i = 0
         while i + 1 < len(bxs):
             b = bxs[i]
@@ -1139,20 +1148,94 @@ class RAGFlowPdfParser:
             need_image, zoomin, return_html, False)
         return self.__filterout_scraps(deepcopy(self.boxes), zoomin), tbls
 
+    def parse_into_bboxes(self, fnm, callback=None, zoomin=3):
+        start = timer()
+        self.__images__(fnm, zoomin)
+        if callback:
+            callback(0.40, "OCR finished ({:.2f}s)".format(timer() - start))
+
+        start = timer()
+        self._layouts_rec(zoomin)
+        if callback:
+            callback(0.63, "Layout analysis ({:.2f}s)".format(timer() - start))
+
+        start = timer()
+        self._table_transformer_job(zoomin)
+        if callback:
+            callback(0.83, "Table analysis ({:.2f}s)".format(timer() - start))
+
+        start = timer()
+        self._text_merge()
+        self._concat_downward()
+        self._naive_vertical_merge(zoomin)
+        if callback:
+            callback(0.92, "Text merged ({:.2f}s)".format(timer() - start))
+
+        start = timer()
+        tbls, figs = self._extract_table_figure(True, zoomin, True, True, True)
+
+        def insert_table_figures(tbls_or_figs, layout_type):
+            def min_rectangle_distance(rect1, rect2):
+                import math
+                pn1, left1, right1, top1, bottom1 = rect1
+                pn2, left2, right2, top2, bottom2 = rect2
+                if (right1 >= left2 and right2 >= left1 and
+                        bottom1 >= top2 and bottom2 >= top1):
+                    return 0 + (pn1-pn2)*10000
+                if right1 < left2:
+                    dx = left2 - right1
+                elif right2 < left1:
+                    dx = left1 - right2
+                else:
+                    dx = 0
+                if bottom1 < top2:
+                    dy = top2 - bottom1
+                elif bottom2 < top1:
+                    dy = top1 - bottom2
+                else:
+                    dy = 0
+                return math.sqrt(dx*dx + dy*dy) + (pn1-pn2)*10000
+
+            for (img, txt), poss in tbls_or_figs:
+                bboxes = [(i, (b["page_number"], b["x0"], b["x1"], b["top"], b["bottom"])) for i, b in enumerate(self.boxes)]
+                dists = [(min_rectangle_distance((pn, left, right, top, bott), rect),i) for i, rect in bboxes for pn, left, right, top, bott in poss]
+                min_i = np.argmin(dists, axis=0)[0]
+                min_i, rect = bboxes[dists[min_i][-1]]
+                if isinstance(txt, list):
+                    txt = "\n".join(txt)
+                self.boxes.insert(min_i, {
+                    "page_number": rect[0], "x0": rect[1], "x1": rect[2], "top": rect[3], "bottom": rect[4], "layout_type": layout_type, "text": txt, "image": img
+                })
+
+        for b in self.boxes:
+            b["position_tag"] = self._line_tag(b, zoomin)
+            b["image"] = self.crop(b["position_tag"], zoomin)
+
+        insert_table_figures(tbls, "table")
+        insert_table_figures(figs, "figure")
+        if callback:
+            callback(1, "Structured ({:.2f}s)".format(timer() - start))
+        return deepcopy(self.boxes)
+
     @staticmethod
     def remove_tag(txt):
         return re.sub(r"@@[\t0-9.-]+?##", "", txt)
 
-    def crop(self, text, ZM=3, need_position=False):
-        imgs = []
+    @staticmethod
+    def extract_positions(txt):
         poss = []
-        for tag in re.findall(r"@@[0-9-]+\t[0-9.\t]+##", text):
+        for tag in re.findall(r"@@[0-9-]+\t[0-9.\t]+##", txt):
             pn, left, right, top, bottom = tag.strip(
                 "#").strip("@").split("\t")
             left, right, top, bottom = float(left), float(
                 right), float(top), float(bottom)
             poss.append(([int(p) - 1 for p in pn.split("-")],
                          left, right, top, bottom))
+        return poss
+
+    def crop(self, text, ZM=3, need_position=False):
+        imgs = []
+        poss = self.extract_positions(text)
         if not poss:
             if need_position:
                 return None, None
@@ -1296,8 +1379,8 @@ class VisionParser(RAGFlowPdfParser):
 
     def __call__(self, filename, from_page=0, to_page=100000, **kwargs):
         callback = kwargs.get("callback", lambda prog, msg: None)
-
-        self.__images__(fnm=filename, zoomin=3, page_from=from_page, page_to=to_page, **kwargs)
+        zoomin = kwargs.get("zoomin", 3)
+        self.__images__(fnm=filename, zoomin=zoomin, page_from=from_page, page_to=to_page, callback=callback)
 
         total_pdf_pages = self.total_page
 
@@ -1311,16 +1394,19 @@ class VisionParser(RAGFlowPdfParser):
             if pdf_page_num < start_page or pdf_page_num >= end_page:
                 continue
 
-            docs = picture_vision_llm_chunk(
+            text = picture_vision_llm_chunk(
                 binary=img_binary,
                 vision_model=self.vision_model,
                 prompt=vision_llm_describe_prompt(page=pdf_page_num+1),
                 callback=callback,
             )
+            if kwargs.get("callback"):
+                kwargs["callback"](idx*1./len(self.page_images), f"Processed: {idx+1}/{len(self.page_images)}")
 
-            if docs:
-                all_docs.append(docs)
-        return [(doc, "") for doc in all_docs], []
+            if text:
+                width, height = self.page_images[idx].size
+                all_docs.append((text, f"{pdf_page_num+1} 0 {width/zoomin} 0 {height/zoomin}"))
+        return all_docs, []
 
 
 if __name__ == "__main__":

deepdoc/vision/seeit.py

@@ -31,11 +31,11 @@ def save_results(image_list, results, labels, output_dir='output/', threshold=0.
         logging.debug("save result to: " + out_path)
 
 
-def draw_box(im, result, lables, threshold=0.5):
+def draw_box(im, result, labels, threshold=0.5):
     draw_thickness = min(im.size) // 320
     draw = ImageDraw.Draw(im)
-    color_list = get_color_map_list(len(lables))
-    clsid2color = {n.lower():color_list[i] for i,n in enumerate(lables)}
+    color_list = get_color_map_list(len(labels))
+    clsid2color = {n.lower():color_list[i] for i,n in enumerate(labels)}
     result = [r for r in result if r["score"] >= threshold]
 
     for dt in result:

docker/.env

@@ -93,13 +93,13 @@ REDIS_PASSWORD=infini_rag_flow
 SVR_HTTP_PORT=9380
 
 # The RAGFlow Docker image to download.
-# Defaults to the v0.20.1-slim edition, which is the RAGFlow Docker image without embedding models.
-RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.1-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:v0.20.1
+# RAGFLOW_IMAGE=infiniflow/ragflow:v0.20.4
 #
-# The Docker image of the v0.20.1 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:v0.20.1-slim` (default): The RAGFlow Docker image without embedding models.  
-  - `infiniflow/ragflow:v0.20.1`: 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/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;

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:v0.20.1-slim` (default): The RAGFlow Docker image without embedding models.  
-  - `infiniflow/ragflow:v0.20.1`: 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:v0.20.1-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:v0.20.1-slim`
-- **Full edition**: includes built-in embedding models and has no suffix added to the version name. Example: `infiniflow/ragflow:v0.20.1`
+- **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`
 
 ---
 
 ### Which embedding models can be deployed locally?
 
-RAGFlow offers two Docker image editions, `v0.20.1-slim` and `v0.20.1`:  
+RAGFlow offers two Docker image editions, `v0.20.4-slim` and `v0.20.4`:  
   
-- `infiniflow/ragflow:v0.20.1-slim` (default): The RAGFlow Docker image without embedding models.  
-- `infiniflow/ragflow:v0.20.1`: 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 v0.20.1 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.

docs/guides/agent/agent_component_reference/retrieval.mdx

@@ -9,19 +9,70 @@ 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 v0.20.1, 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. A **Retrieval** component can operate either as a standalone workflow module or as a tool for an **Agent** component. In the latter role, the **Agent** component has autonomous control over when to invoke it for query and retrieval.
+
+The following screenshot shows a reference design using the **Retrieval** component, where the component serves as a tool for an **Agent** component. You can find it from the **Report Agent Using Knowledge Base** Agent template.
+
+![retrieval_reference_design](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/retrieval_reference_design.jpg)
+
+## Prerequisites
+
+Ensure you [have properly configured your target knowledge base(s)](../../dataset/configure_knowledge_base.md).
+
+## Quickstart
+
+### 1. Click on a **Retrieval** component to show its configuration panel  
+
+The corresponding configuration panel appears to the right of the canvas. Use this panel to define and fine-tune the **Retrieval** component's search behavior.
+
+### 2. Input query variable(s)
+
+The **Retrieval** component relies on query variables to specify its queries. 
+
+:::caution IMPORTANT
+- If you use the **Retrieval** component as a standalone workflow module, input query variables in the **Input Variables** text box.
+- If it is used as a tool for an **Agent** component, input the query variables in the **Agent** component's **User prompt** field.
+:::
+
+By default, you can use `sys.query`, which is the user query and the default output of the **Begin** component. All global variables defined before the **Retrieval** component can also be used as query statements. Use the `(x)` button or type `/` to show all the available query variables.
+
+### 3. Select knowledge base(s) to query
+
+You can specify one or multiple knowledge bases to retrieve data from. If selecting mutiple, ensure they use the same embedding model.
+
+### 4. Expand **Advanced Settings** to configure the retrieval method
+
+By default, a combination of weighted keyword similarity and weighted vector cosine similarity is used during retrieval. If a rerank model is selected, a combination of weighted keyword similarity and weighted reranking score will be used for retrieval.
+
+As a starter, you can skip this step to stay with the default retrieval method.
+
+:::caution WARNING
+Using a rerank model will *significantly* increase the system's response time. If you must use a rerank model, ensure you use a SaaS reranker; if you prefer a locally deployed rerank model, ensure you start RAGFlow with **docker-compose-gpu.yml**.
+:::
+
+### 5. Enable cross-language search
+
+If your user query is different from the languages of the knowledge bases, you can select the target languages in the **Cross-language search** dropdown menu. The model will then translates queries to ensure accurate matching of semantic meaning across languages.
+
+
+### 6. Test retrieval results
+
+Click the triangle button on the top of canvas to test the retrieval results.
+
+### 6. Choose the next component
+
+When necessary, click the **+** button on the **Retrieval** component to choose the next component in the worflow from the dropdown list.
+
 
 ## Configurations
 
-Click on a **Retrieval** component to open its configuration window.
-
 ### Query variables
 
 *Mandatory*
 
-Select the query source for retrieval.
+Select the query source for retrieval. Defaults to `sys.query`, which is the default output of the **Begin** component.
 
-The **Retrieval** component relies on query variables to specify its data inputs (queries). All global variables defined before the **Retrieval** component are available in the dropdown list.  
+The **Retrieval** component relies on query variables to specify its queries. All global variables defined before the **Retrieval** component can also be used as queries. Use the `(x)` button or type `/` to show all the available query variables.
 
 ### Knowledge bases 
 
@@ -72,8 +123,23 @@ Select one or more languages for cross‑language search. If no language is sele
 
 ### Use knowledge graph
 
+:::caution IMPORTANT
+Before enabling this feature, ensure you have properly [constructed a knowledge graph from each target knowledge base](../../dataset/construct_knowledge_graph.md).
+:::
+
 Whether to use knowledge graph(s) in the specified knowledge base(s) during retrieval for multi-hop question answering. When enabled, this would involve iterative searches across entity, relationship, and community report chunks, greatly increasing retrieval time.
 
 ### Output
 
 The global variable name for the output of the **Retrieval** component, which can be referenced by other components in the workflow.
+
+
+## Frequently asked questions
+
+### How to reduce response time?
+
+Go through the checklist below for best performance:
+
+- Leave the **Rerank model** field empty.
+- If you must use a rerank model, ensure you use a SaaS reranker; if you prefer a locally deployed rerank model, ensure you start RAGFlow with **docker-compose-gpu.yml**.
+- Disable **Use knowledge graph**.

docs/guides/agent/agent_introduction.md

@@ -9,12 +9,12 @@ Key concepts, basic operations, a quick view of the agent editor.
 
 ---
 
-## Key concepts
-
 :::danger DEPRECATED!
 A new version is coming soon.
 :::
 
+## Key concepts
+
 Agents and RAG are complementary techniques, each enhancing the other’s capabilities in business applications. RAGFlow v0.8.0 introduces an agent mechanism, featuring a no-code workflow editor on the front end and a comprehensive graph-based task orchestration framework on the back end. This mechanism is built on top of RAGFlow's existing RAG solutions and aims to orchestrate search technologies such as query intent classification, conversation leading, and query rewriting to:
 
 - Provide higher retrievals and,
@@ -33,55 +33,19 @@ Before proceeding, ensure that:
 
 Click the **Agent** tab in the middle top of the page to show the **Agent** page. As shown in the screenshot below, the cards on this page represent the created agents, which you can continue to edit.
 
-![agent_mainpage](https://github.com/user-attachments/assets/5c0bb123-8f4e-42ea-b250-43f640dc6814)
+![Agent_list](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/agent_list.jpg)
 
 We also provide templates catered to different business scenarios. You can either generate your agent from one of our agent templates or create one from scratch:
 
 1. Click **+ Create agent** to show the **agent template** page:
 
-   ![agent_templates](https://github.com/user-attachments/assets/73bd476c-4bab-4c8c-82f8-6b00fb2cd044)
+   ![agent_template](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/agent_template_list.jpg)
 
-2. To create an agent from scratch, click the **Blank** card. Alternatively, to create an agent from one of our templates, hover over the desired card, such as **General-purpose chatbot**, click **Use this template**, name your agent in the pop-up dialogue, and click **OK** to confirm.  
+2. To create an agent from scratch, click **Create Agent**. Alternatively, to create an agent from one of our templates, click the desired card, such as **Deep Research**, name your agent in the pop-up dialogue, and click **OK** to confirm.  
 
-   *You are now taken to the **no-code workflow editor** page. The left panel lists the components (operators): Above the dividing line are the RAG-specific components; below the line are tools. We are still working to expand the component list.*
+   *You are now taken to the **no-code workflow editor** page.*
 
-   ![workflow_editor](https://github.com/user-attachments/assets/47b4d5ce-b35a-4d6b-b483-ba495a75a65d)
+   ![add_component](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/add_component.jpg)
 
-3. General speaking, now you can do the following:
-   - Drag and drop a desired component to your workflow,
-   - Select the knowledge base to use,
-   - Update settings of specific components,
-   - Update LLM settings
-   - Sets the input and output for a specific component, and more.
-4. Click **Save** to apply changes to your agent and **Run** to test it.
-
-## Components
-
-Please review the flowing description of the RAG-specific components before you proceed:
-
-| Component      | Description                                                                                                                                                                                              |
-|----------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
-| **Retrieval**  | A component that retrieves information from specified knowledge bases and returns 'Empty response' if no information is found. Ensure the correct knowledge bases are selected.                          |
-| **Generate**   | A component that prompts the LLM to generate responses. You must ensure the prompt is set correctly.                                                                                                     |
-| **Interact**   | A component that serves as the interface between human and the bot, receiving user inputs and displaying the agent's responses.                                                                          |
-| **Categorize** | A component that uses the LLM to classify user inputs into predefined categories. Ensure you specify the name, description, and examples for each category, along with the corresponding next component. |
-| **Message**    | A component that sends out a static message. If multiple messages are supplied, it randomly selects one to send. Ensure its downstream is **Interact**, the interface component.                         |
-| **Rewrite**    | A component that rewrites a user query from the **Interact** component, based on the context of previous dialogues.                                                                                      |
-| **Keyword**    | A component that extracts keywords from a user query, with TopN specifying the number of keywords to extract.                                                                                            |
-
-:::caution NOTE
-
-- Ensure **Rewrite**'s upstream component is **Relevant** and downstream component is **Retrieval**.
-- Ensure the downstream component of **Message** is **Interact**.
-- The downstream component of **Begin** is always **Interact**.
-
-:::
-
-## Basic operations
-
-| Operation                 | Description                                                                                                                              |
-|---------------------------|------------------------------------------------------------------------------------------------------------------------------------------|
-| Add a component           | Drag and drop the desired component from the left panel onto the canvas.                                                                 |
-| Delete a component        | On the canvas, hover over the three dots (...) of the component to display the delete option, then select it to remove the component.    |
-| Copy a component          | On the canvas, hover over the three dots (...) of the component to display the copy option, then select it to make a copy the component. |
-| Update component settings | On the canvas, click the desired component to display the component settings.                                                            |
+3. Click the **+** button on the **Begin** component to select the desired components in your workflow.
+4. Click **Save** to apply changes to your agent.

docs/guides/agent/embed_agent_into_webpage.md

@@ -10,4 +10,6 @@ You can use iframe to embed an agent into a third-party webpage.
 1. Before proceeding, you must [acquire an API key](../models/llm_api_key_setup.md); otherwise, an error message would appear.
 2. On the **Agent** page, click an intended agent to access its editing page.
 3. Click **Management > Embed into webpage** on the top right corner of the canvas to show the **iframe** window:
-4. Copy the iframe and embed it into a specific location on your webpage.
+4. Copy the iframe and embed it into your webpage.
+
+![Embed_agent](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/embed_agent_into_webpage.jpg)

docs/guides/agent/general_purpose_chatbot.md

@@ -1,109 +0,0 @@
----
-sidebar_position: 2
-slug: /general_purpose_chatbot
----
-
-# Create chatbot
-
-Create a general-purpose chatbot.
-
----
-
-:::danger DEPRECATED!
-A new version is coming soon.
-:::
-
-Chatbot is one of the most common AI scenarios. However, effectively understanding user queries and responding appropriately remains a challenge. RAGFlow's general-purpose chatbot agent is our attempt to tackle this longstanding issue.  
-
-This chatbot closely resembles the chatbot introduced in [Start an AI chat](../chat/start_chat.md), but with a key difference - it introduces a reflective mechanism that allows it to improve the retrieval from the target knowledge bases by rewriting the user's query.
-
-This document provides guides on creating such a chatbot using our chatbot template.
-
-## Prerequisites
-
-1. Ensure you have properly set the LLM to use. See the guides on [Configure your API key](../models/llm_api_key_setup.md) or [Deploy a local LLM](../models/deploy_local_llm.mdx) for more information.
-2. Ensure you have a knowledge base configured and the corresponding files properly parsed. See the guide on [Configure a knowledge base](../dataset/configure_knowledge_base.md) for more information.
-3. Make sure you have read the [Introduction to Agentic RAG](./agent_introduction.md).
-
-## Create a chatbot agent from template
-
-To create a general-purpose chatbot agent using our template:
-
-1. Click the **Agent** tab in the middle top of the page to show the **Agent** page.
-2. Click **+ Create agent** on the top right of the page to show the **agent template** page.
-3. On the **agent template** page, hover over the card on **General-purpose chatbot** and click **Use this template**.  
-   *You are now directed to the **no-code workflow editor** page.*
-
-   ![workflow_editor](https://github.com/user-attachments/assets/52e7dc62-4bf5-4fbb-ab73-4a6e252065f0)
-
-:::tip NOTE
-RAGFlow's no-code editor spares you the trouble of coding, making agent development effortless.
-:::
-
-## Understand each component in the template
-
-Here’s a breakdown of each component and its role and requirements in the chatbot template:
-
-- **Begin**
-  - Function: Sets an opening greeting for users.
-  - Purpose: Establishes a welcoming atmosphere and prepares the user for interaction.
-
-- **Interact**
-  - Function: Serves as the interface between human and the bot.
-  - Role: Acts as the downstream component of **Begin**.  
-
-- **Retrieval**
-  - Function: Retrieves information from specified knowledge base(s).
-  - Requirement: Must have `knowledgebases` set up to function.
-
-- **Relevant**
-  - Function: Assesses the relevance of the retrieved information from the **Retrieval** component to the user query.
-  - Process:  
-    - If relevant, it directs the data to the **Generate** component for final response generation.
-    - Otherwise, it triggers the **Rewrite** component to refine the user query and redo the retrival process.
-
-- **Generate**
-  - Function: Prompts the LLM to generate responses based on the retrieved information.  
-  - Note: The prompt settings allow you to control the way in which the LLM generates responses. Be sure to review the prompts and make necessary changes.
-
-- **Rewrite**:  
-  - Function: Refines a user query when no relevant information from the knowledge base is retrieved.  
-  - Usage: Often used in conjunction with **Relevant** and **Retrieval** to create a reflective/feedback loop.  
-
-## Configure your chatbot agent
-
-1. Click **Begin** to set an opening greeting:  
-   ![opener](https://github.com/user-attachments/assets/4416bc16-2a84-4f24-a19b-6dc8b1de0908)
-
-2. Click **Retrieval** to select the right knowledge base(s) and make any necessary adjustments:  
-   ![setting_knowledge_bases](https://github.com/user-attachments/assets/5f694820-5651-45bc-afd6-cf580ca0228d)
-
-3. Click **Generate** to configure the LLM's summarization behavior:  
-   3.1. Confirm the model.  
-   3.2. Review the prompt settings. If there are variables, ensure they match the correct component IDs:  
-   ![prompt_settings](https://github.com/user-attachments/assets/19e94ea7-7f62-4b73-b526-32fcfa62f1e9)
-
-4. Click **Relevant** to review or change its settings:  
-   *You may retain the current settings, but feel free to experiment with changes to understand how the agent operates.*
-   ![relevant_settings](https://github.com/user-attachments/assets/9ff7fdd8-7a69-4ee2-bfba-c7fb8029150f)
-
-5. Click **Rewrite** to select a different model for query rewriting or update the maximum loop times for query rewriting:  
-   ![choose_model](https://github.com/user-attachments/assets/2bac1d6c-c4f1-42ac-997b-102858c3f550)
-   ![loop_time](https://github.com/user-attachments/assets/09a4ce34-7aac-496f-aa59-d8aa33bf0b1f)
-
-:::danger NOTE
-Increasing the maximum loop times may significantly extend the time required to receive the final response.
-:::
-
-1. Update your workflow where you see necessary.
-
-2. Click to **Save** to apply your changes.  
-   *Your agent appears as one of the agent cards on the **Agent** page.*
-
-## Test your chatbot agent
-
-1. Find your chatbot agent on the **Agent** page:  
-   ![find_chatbot](https://github.com/user-attachments/assets/6e6382c6-9a86-4190-9fdd-e363b7f64ba9)
-
-2. Experiment with your questions to verify if this chatbot functions as intended:  
-   ![test_chatbot](https://github.com/user-attachments/assets/c074d3bd-4c39-4b05-a68b-1fd361f256b3)

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/ai_search.md

@@ -11,7 +11,9 @@ Conduct an AI search.
 
 An AI search is a single-turn AI conversation using a predefined retrieval strategy (a hybrid search of weighted keyword similarity and weighted vector similarity) and the system's default chat model. It does not involve advanced RAG strategies like knowledge graph, auto-keyword, or auto-question. The related chunks are listed below the chat model's response in descending order based on their similarity scores. 
 
-![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/ai_search.jpg)
+![Create search app](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/create_search_app.jpg)
+
+![Search view](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/search_view.jpg)
 
 :::tip NOTE
 When debugging your chat assistant, you can use AI search as a reference to verify your model settings and retrieval strategy.
@@ -22,10 +24,8 @@ When debugging your chat assistant, you can use AI search as a reference to veri
 - Ensure that you have configured the system's default models on the **Model providers** page.
 - Ensure that the intended knowledge bases are properly configured and the intended documents have finished file parsing.
 
-
 ## Frequently asked questions
 
 ### Key difference between an AI search and an AI chat?
 
 A chat is a multi-turn AI conversation where you can define your retrieval strategy (a weighted reranking score can be used to replace the weighted vector similarity in a hybrid search) and choose your chat model. In an AI chat, you can configure advanced RAG strategies, such as knowledge graphs, auto-keyword, and auto-question, for your specific case. Retrieved chunks are not displayed along with the answer.
-

docs/guides/chat/implement_deep_research.md

@@ -15,13 +15,13 @@ From v0.17.0 onward, RAGFlow supports integrating agentic reasoning in an AI cha
 
 To activate this feature:
 
-1. Enable the **Reasoning** toggle under the **Prompt engine** tab of your chat assistant dialogue.
+1. Enable the **Reasoning** toggle in **Chat setting**.
 
-![Image](https://github.com/user-attachments/assets/4a1968d0-0128-4371-879f-77f3a70197f5)
+![chat_reasoning](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/chat_reasoning.jpg)
 
-2. Enter the correct Tavily API key under the **Assistant settings** tab of your chat assistant dialogue to leverage Tavily-based web search
+2. Enter the correct Tavily API key to leverage Tavily-based web search:
 
-![Image](https://github.com/user-attachments/assets/e8787532-7e72-49ef-8951-169ae544512f)
+![chat_tavily](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/chat_tavily.jpg)
 
 *The following is a screenshot of a conversation that integrates Deep Research:*
 

docs/guides/chat/set_chat_variables.md

@@ -9,7 +9,7 @@ Set variables to be used together with the system prompt for your LLM.
 
 ---
 
-When configuring the system prompt for a chat model, variables play an important role in enhancing flexibility and reusability. With variables, you can dynamically adjust the system prompt to be sent to your model. In the context of RAGFlow, if you have defined variables in the **Chat Configuration** dialogue, except for the system's reserved variable `{knowledge}`, you are required to pass in values for them from RAGFlow's [HTTP API](../../references/http_api_reference.md#converse-with-chat-assistant) or through its [Python SDK](../../references/python_api_reference.md#converse-with-chat-assistant).
+When configuring the system prompt for a chat model, variables play an important role in enhancing flexibility and reusability. With variables, you can dynamically adjust the system prompt to be sent to your model. In the context of RAGFlow, if you have defined variables in **Chat setting**, except for the system's reserved variable `{knowledge}`, you are required to pass in values for them from RAGFlow's [HTTP API](../../references/http_api_reference.md#converse-with-chat-assistant) or through its [Python SDK](../../references/python_api_reference.md#converse-with-chat-assistant).
 
 :::danger IMPORTANT
 In RAGFlow, variables are closely linked with the system prompt. When you add a variable in the **Variable** section, include it in the system prompt. Conversely, when deleting a variable, ensure it is removed from the system prompt; otherwise, an error would occur.
@@ -17,9 +17,7 @@ In RAGFlow, variables are closely linked with the system prompt. When you add a
 
 ## Where to set variables
 
-Hover your mouse over your chat assistant, click **Edit** to open its **Chat Configuration** dialogue, then click the **Prompt engine** tab. Here, you can work on your variables in the **System prompt** field and the **Variable** section:
-
-![set_variables](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/prompt_engine.jpg)
+![set_variables](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/chat_variables.jpg)
 
 ## 1. Manage variables
 
@@ -42,8 +40,6 @@ Besides `{knowledge}`, you can also define your own variables to pair with the s
 - **Disabled** (Default): The variable is mandatory and must be provided.
 - **Enabled**: The variable is optional and can be omitted if not needed.
 
-
-
 ## 2. Update system prompt
 
 After you add or remove variables in the **Variable** section, ensure your changes are reflected in the system prompt to avoid inconsistencies or errors. Here's an example:

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 v0.20.1, 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).
 
@@ -77,28 +77,24 @@ You start an AI conversation by creating an assistant.
 
 5. Now, let's start the show:
 
-   ![question1](https://github.com/user-attachments/assets/c4114a3d-74ff-40a3-9719-6b47c7b11ab1)
+   ![chat_thermal_solution](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/chat_thermal_solution.jpg)
 
 :::tip NOTE
 
 1. Click the light bulb icon above the answer to view the expanded system prompt:
 
-![](https://github.com/user-attachments/assets/515ab187-94e8-412a-82f2-aba52cd79e09)
+![prompt_display](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/prompt_display.jpg)
 
    *The light bulb icon is available only for the current dialogue.*
 
 2. Scroll down the expanded prompt to view the time consumed for each task:
 
-![enlighten](https://github.com/user-attachments/assets/fedfa2ee-21a7-451b-be66-20125619923c)
+![time_elapsed](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/time_elapsed.jpg)
 :::
 
 ## Update settings of an existing chat assistant
 
-Hover over an intended chat assistant **>** **Edit** to show the chat configuration dialogue:
-
-![edit_chat](https://github.com/user-attachments/assets/5c514cf0-a959-4cfe-abad-5e42a0e23974)
-
-![chat_config](https://github.com/user-attachments/assets/1a4eaed2-5430-4585-8ab6-930549838c5b)
+![chat_setting](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/chat_setting.jpg)
 
 ## Integrate chat capabilities into your application or webpage
 
@@ -113,6 +109,8 @@ You can use iframe to embed the created chat assistant into a third-party webpag
 1. Before proceeding, you must [acquire an API key](../models/llm_api_key_setup.md); otherwise, an error message would appear.
 2. Hover over an intended chat assistant **>** **Edit** to show the **iframe** window:
 
-   ![chat-embed](https://github.com/user-attachments/assets/13ea3021-31c4-4a14-9b32-328cd3318fb5)
+   ![chat-embed](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/embed_chat_into_webpage.jpg)
 
-3. Copy the iframe and embed it into a specific location on your webpage.
+3. Copy the iframe and embed it into your webpage.
+
+![chat-embed](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/embedded_chat_app.jpg)

docs/guides/dataset/configure_knowledge_base.md

@@ -16,7 +16,7 @@ Knowledge base, hallucination-free chat, and file management are the three pilla
 
 With multiple knowledge bases, you can build more flexible, diversified question answering. To create your first knowledge base:
 
-![create knowledge base](https://github.com/infiniflow/ragflow/assets/93570324/110541ed-6cea-4a03-a11c-414a0948ba80)
+![create knowledge base](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/create_knowledge_base.jpg)
 
 _Each time a knowledge base is created, a folder with the same name is generated in the **root/.knowledgebase** directory._
 
@@ -24,7 +24,7 @@ _Each time a knowledge base is created, a folder with the same name is generated
 
 The following screenshot shows the configuration page of a knowledge base. A proper configuration of your knowledge base is crucial for future AI chats. For example, choosing the wrong embedding model or chunking method would cause unexpected semantic loss or mismatched answers in chats. 
 
-![knowledge base configuration](https://github.com/infiniflow/ragflow/assets/93570324/384c671a-8b9c-468c-b1c9-1401128a9b65)
+![knowledge base configuration](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/configure_knowledge_base.jpg)
 
 This section covers the following topics:
 
@@ -56,7 +56,7 @@ RAGFlow offers multiple chunking template to facilitate chunking files of differ
 
 You can also change a file's chunking method on the **Datasets** page.
 
-![change chunking method](https://github.com/infiniflow/ragflow/assets/93570324/ac116353-2793-42b2-b181-65e7082bed42)
+![change chunking method](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/embedded_chat_app.jpg)
 
 ### Select embedding model
 
@@ -82,10 +82,8 @@ While uploading files directly to a knowledge base seems more convenient, we *hi
 
 File parsing is a crucial topic in knowledge base configuration. The meaning of file parsing in RAGFlow is twofold: chunking files based on file layout and building embedding and full-text (keyword) indexes on these chunks. After having selected the chunking method and embedding model, you can start parsing a file:
 
-![parse file](https://github.com/infiniflow/ragflow/assets/93570324/5311f166-6426-447f-aa1f-bd488f1cfc7b)
+![parse file](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/parse_file.jpg)
 
-- Click the play button next to **UNSTART** to start file parsing.
-- Click the red-cross icon and then refresh, if your file parsing stalls for a long time. 
 - As shown above, RAGFlow allows you to use a different chunking method for a particular file, offering flexibility beyond the default method. 
 - As shown above, RAGFlow allows you to enable or disable individual files, offering finer control over knowledge base-based AI chats. 
 
@@ -97,13 +95,13 @@ RAGFlow features visibility and explainability, allowing you to view the chunkin
 
    _You are taken to the **Chunk** page:_
 
-   ![chunks](https://github.com/infiniflow/ragflow/assets/93570324/0547fd0e-e71b-41f8-8e0e-31649c85fd3d)
+   ![chunks](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/file_chunks.jpg)
 
 2. Hover over each snapshot for a quick view of each chunk.
 
-3. Double-click the chunked texts to add keywords or make *manual* changes where necessary:
+3. Double-click the chunked texts to add keywords, questions, tags, or make *manual* changes where necessary:
 
-   ![update chunk](https://github.com/infiniflow/ragflow/assets/93570324/1d84b408-4e9f-46fd-9413-8c1059bf9c76)
+   ![update chunk](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/add_keyword_question.jpg)
 
 :::caution NOTE
 You can add keywords to a file chunk to increase its ranking for queries containing those keywords. This action increases its keyword weight and can improve its position in search list.  
@@ -113,7 +111,7 @@ You can add keywords to a file chunk to increase its ranking for queries contain
 
    _As you can tell from the following, RAGFlow responds with truthful citations._
 
-   ![retrieval test](https://github.com/infiniflow/ragflow/assets/93570324/c03f06f6-f41f-4b20-a97e-ae405d3a950c)
+   ![retrieval test](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/retrieval_test.jpg)
 
 ### Run retrieval testing
 
@@ -124,13 +122,11 @@ RAGFlow uses multiple recall of both full-text search and vector search in its c
 
 See [Run retrieval test](./run_retrieval_test.md) for details.
 
-![retrieval test](https://github.com/infiniflow/ragflow/assets/93570324/c03f06f6-f41f-4b20-a97e-ae405d3a950c)
-
 ## Search for knowledge base
 
-As of RAGFlow v0.20.1, 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)
+![search knowledge base](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/search_datasets.jpg)
 
 ## Delete knowledge base
 
@@ -139,4 +135,4 @@ You are allowed to delete a knowledge base. Hover your mouse over the three dot
 - The files uploaded directly to the knowledge base are gone;  
 - The file references, which you created from within **File Management**, are gone, but the associated files still exist in **File Management**. 
 
-![delete knowledge base](https://github.com/infiniflow/ragflow/assets/93570324/fec7a508-6cfe-4bca-af90-81d3fdb94098)
+![delete knowledge base](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/delete_datasets.jpg)

docs/guides/dataset/construct_knowledge_graph.md

@@ -31,7 +31,7 @@ RAPTOR (Recursive Abstractive Processing for Tree Organized Retrieval) can also
 
 The system's default chat model is used to generate knowledge graph. Before proceeding, ensure that you have a chat model properly configured:
 
-![Image](https://github.com/user-attachments/assets/6bc34279-68c3-4d99-8d20-b7bd1dafc1c1)
+![Set default models](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/set_default_models.jpg)
 
 ## Configurations
 
@@ -74,7 +74,7 @@ In a knowledge graph, a community is a cluster of entities linked by relationshi
 3. Click **Knowledge graph** to view the details of the generated graph.
 4. To use the created knowledge graph, do either of the following:
    
-   - In your **Chat Configuration** dialogue, click the **Assistant settings** tab to add the corresponding knowledge base(s) and click the **Prompt engine** tab to switch on the **Use knowledge graph** toggle.
+   - In the **Chat setting** panel of your chat app, switch on the **Use knowledge graph** toggle.
    - If you are using an agent, click the **Retrieval** agent component to specify the knowledge base(s) and switch on the **Use knowledge graph** toggle.
 
 ## Frequently asked questions