Docs: What is RAG & What is Agent context engine (#12341)

### Type of change

- [x] Documentation Update

.github/workflows/release.yml

@@ -10,6 +10,12 @@ on:
     tags:
       - "v*.*.*"                  # normal release
 
+permissions:
+  contents: write
+  actions: read
+  checks: read
+  statuses: read
+
 # https://docs.github.com/en/actions/using-jobs/using-concurrency
 concurrency:
   group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
@@ -76,6 +82,14 @@ jobs:
           # The body field does not support environment variable substitution directly.
           body_path: release_body.md
 
+      - name: Build and push image
+        run: |
+          sudo docker login --username infiniflow --password-stdin <<< ${{ secrets.DOCKERHUB_TOKEN }}
+          sudo docker build --build-arg NEED_MIRROR=1 --build-arg HTTPS_PROXY=${HTTPS_PROXY} --build-arg HTTP_PROXY=${HTTP_PROXY} -t infiniflow/ragflow:${RELEASE_TAG} -f Dockerfile .
+          sudo docker tag infiniflow/ragflow:${RELEASE_TAG} infiniflow/ragflow:latest
+          sudo docker push infiniflow/ragflow:${RELEASE_TAG}
+          sudo docker push infiniflow/ragflow:latest
+
       - name: Build and push ragflow-sdk
         if: startsWith(github.ref, 'refs/tags/v')
         run: |
@@ -85,11 +99,3 @@ jobs:
         if: startsWith(github.ref, 'refs/tags/v')
         run: |
           cd admin/client && uv build && uv publish --token ${{ secrets.PYPI_API_TOKEN }}
-
-      - name: Build and push image
-        run: |
-          sudo docker login --username infiniflow --password-stdin <<< ${{ secrets.DOCKERHUB_TOKEN }}
-          sudo docker build --build-arg NEED_MIRROR=1 --build-arg HTTPS_PROXY=${HTTPS_PROXY} --build-arg HTTP_PROXY=${HTTP_PROXY} -t infiniflow/ragflow:${RELEASE_TAG} -f Dockerfile .
-          sudo docker tag infiniflow/ragflow:${RELEASE_TAG} infiniflow/ragflow:latest
-          sudo docker push infiniflow/ragflow:${RELEASE_TAG}
-          sudo docker push infiniflow/ragflow:latest

.github/workflows/tests.yml

@@ -205,29 +205,29 @@ jobs:
       - name: Run sdk tests against Elasticsearch
         run: |
           export http_proxy=""; export https_proxy=""; export no_proxy=""; export HTTP_PROXY=""; export HTTPS_PROXY=""; export NO_PROXY=""
-          until sudo docker exec ${RAGFLOW_CONTAINER} curl -s --connect-timeout 5 ${HOST_ADDRESS} > /dev/null; do
+          until sudo docker exec ${RAGFLOW_CONTAINER} curl -s --connect-timeout 5 ${HOST_ADDRESS}/v1/system/ping > /dev/null; do
             echo "Waiting for service to be available..."
             sleep 5
           done
-          source .venv/bin/activate && pytest -s --tb=short --level=${HTTP_API_TEST_LEVEL} test/testcases/test_sdk_api 2>&1 | tee es_sdk_test.log
+          source .venv/bin/activate && set -o pipefail; pytest -s --tb=short --level=${HTTP_API_TEST_LEVEL} test/testcases/test_sdk_api 2>&1 | tee es_sdk_test.log
 
       - name: Run frontend api tests against Elasticsearch
         run: |
           export http_proxy=""; export https_proxy=""; export no_proxy=""; export HTTP_PROXY=""; export HTTPS_PROXY=""; export NO_PROXY=""
-          until sudo docker exec ${RAGFLOW_CONTAINER} curl -s --connect-timeout 5 ${HOST_ADDRESS} > /dev/null; do
+          until sudo docker exec ${RAGFLOW_CONTAINER} curl -s --connect-timeout 5 ${HOST_ADDRESS}/v1/system/ping > /dev/null; do
             echo "Waiting for service to be available..."
             sleep 5
           done
-          source .venv/bin/activate && pytest -s --tb=short sdk/python/test/test_frontend_api/get_email.py sdk/python/test/test_frontend_api/test_dataset.py 2>&1 | tee es_api_test.log
+          source .venv/bin/activate && set -o pipefail; pytest -s --tb=short sdk/python/test/test_frontend_api/get_email.py sdk/python/test/test_frontend_api/test_dataset.py 2>&1 | tee es_api_test.log
           
       - name: Run http api tests against Elasticsearch
         run: |
           export http_proxy=""; export https_proxy=""; export no_proxy=""; export HTTP_PROXY=""; export HTTPS_PROXY=""; export NO_PROXY=""
-          until sudo docker exec ${RAGFLOW_CONTAINER} curl -s --connect-timeout 5 ${HOST_ADDRESS} > /dev/null; do
+          until sudo docker exec ${RAGFLOW_CONTAINER} curl -s --connect-timeout 5 ${HOST_ADDRESS}/v1/system/ping > /dev/null; do
             echo "Waiting for service to be available..."
             sleep 5
           done
-          source .venv/bin/activate && pytest -s --tb=short --level=${HTTP_API_TEST_LEVEL} test/testcases/test_http_api 2>&1 | tee es_http_api_test.log
+          source .venv/bin/activate && set -o pipefail; pytest -s --tb=short --level=${HTTP_API_TEST_LEVEL} test/testcases/test_http_api 2>&1 | tee es_http_api_test.log
 
       - name: Stop ragflow:nightly
         if: always()  # always run this step even if previous steps failed
@@ -243,29 +243,29 @@ jobs:
       - name: Run sdk tests against Infinity
         run: |
           export http_proxy=""; export https_proxy=""; export no_proxy=""; export HTTP_PROXY=""; export HTTPS_PROXY=""; export NO_PROXY=""
-          until sudo docker exec ${RAGFLOW_CONTAINER} curl -s --connect-timeout 5 ${HOST_ADDRESS} > /dev/null; do
+          until sudo docker exec ${RAGFLOW_CONTAINER} curl -s --connect-timeout 5 ${HOST_ADDRESS}/v1/system/ping > /dev/null; do
             echo "Waiting for service to be available..."
             sleep 5
           done
-          source .venv/bin/activate && DOC_ENGINE=infinity pytest -s --tb=short --level=${HTTP_API_TEST_LEVEL} test/testcases/test_sdk_api 2>&1 | tee infinity_sdk_test.log
+          source .venv/bin/activate && set -o pipefail; DOC_ENGINE=infinity pytest -s --tb=short --level=${HTTP_API_TEST_LEVEL} test/testcases/test_sdk_api 2>&1 | tee infinity_sdk_test.log
 
       - name: Run frontend api tests against Infinity
         run: |
           export http_proxy=""; export https_proxy=""; export no_proxy=""; export HTTP_PROXY=""; export HTTPS_PROXY=""; export NO_PROXY=""
-          until sudo docker exec ${RAGFLOW_CONTAINER} curl -s --connect-timeout 5 ${HOST_ADDRESS} > /dev/null; do
+          until sudo docker exec ${RAGFLOW_CONTAINER} curl -s --connect-timeout 5 ${HOST_ADDRESS}/v1/system/ping > /dev/null; do
             echo "Waiting for service to be available..."
             sleep 5
           done
-          source .venv/bin/activate && DOC_ENGINE=infinity pytest -s --tb=short sdk/python/test/test_frontend_api/get_email.py sdk/python/test/test_frontend_api/test_dataset.py 2>&1 | tee infinity_api_test.log
+          source .venv/bin/activate && set -o pipefail; DOC_ENGINE=infinity pytest -s --tb=short sdk/python/test/test_frontend_api/get_email.py sdk/python/test/test_frontend_api/test_dataset.py 2>&1 | tee infinity_api_test.log
 
       - name: Run http api tests against Infinity
         run: |
           export http_proxy=""; export https_proxy=""; export no_proxy=""; export HTTP_PROXY=""; export HTTPS_PROXY=""; export NO_PROXY=""
-          until sudo docker exec ${RAGFLOW_CONTAINER} curl -s --connect-timeout 5 ${HOST_ADDRESS} > /dev/null; do
+          until sudo docker exec ${RAGFLOW_CONTAINER} curl -s --connect-timeout 5 ${HOST_ADDRESS}/v1/system/ping > /dev/null; do
             echo "Waiting for service to be available..."
             sleep 5
           done
-          source .venv/bin/activate && DOC_ENGINE=infinity pytest -s --tb=short --level=${HTTP_API_TEST_LEVEL} test/testcases/test_http_api 2>&1 | tee infinity_http_api_test.log
+          source .venv/bin/activate && set -o pipefail; DOC_ENGINE=infinity pytest -s --tb=short --level=${HTTP_API_TEST_LEVEL} test/testcases/test_http_api 2>&1 | tee infinity_http_api_test.log
 
       - name: Stop ragflow:nightly
         if: always()  # always run this step even if previous steps failed

Dockerfile

@@ -192,6 +192,7 @@ COPY pyproject.toml uv.lock ./
 COPY mcp mcp
 COPY plugin plugin
 COPY common common
+COPY memory memory
 
 COPY docker/service_conf.yaml.template ./conf/service_conf.yaml.template
 COPY docker/entrypoint.sh ./

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:v0.22.1">
+        <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:v0.23.0">
     </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">
@@ -85,6 +85,7 @@ Try our demo at [https://demo.ragflow.io](https://demo.ragflow.io).
 
 ## 🔥 Latest Updates
 
+- 2025-12-26 Supports 'Memory' for AI agent.
 - 2025-11-19 Supports Gemini 3 Pro.
 - 2025-11-12 Supports data synchronization from Confluence, S3, Notion, Discord, Google Drive.
 - 2025-10-23 Supports MinerU & Docling as document parsing methods.
@@ -187,12 +188,12 @@ 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 `v0.22.1` edition of the RAGFlow Docker image. See the following table for descriptions of different RAGFlow editions. To download a RAGFlow edition different from `v0.22.1`, update the `RAGFLOW_IMAGE` variable accordingly in **docker/.env** before using `docker compose` to start the server.
+> The command below downloads the `v0.23.0` edition of the RAGFlow Docker image. See the following table for descriptions of different RAGFlow editions. To download a RAGFlow edition different from `v0.23.0`, update the `RAGFLOW_IMAGE` variable accordingly in **docker/.env** before using `docker compose` to start the server.
 
 ```bash
    $ cd ragflow/docker
   
-   # git checkout v0.22.1
+   # git checkout v0.23.0
    # Optional: use a stable tag (see releases: https://github.com/infiniflow/ragflow/releases)
    # This step ensures the **entrypoint.sh** file in the code matches the Docker image version.
    
@@ -232,7 +233,7 @@ releases! 🌟
     * Running on all addresses (0.0.0.0)
    ```
 
-   > If you skip this confirmation step and directly log in to RAGFlow, your browser may prompt a `network anormal`
+   > If you skip this confirmation step and directly log in to RAGFlow, your browser may prompt a `network abnormal`
    > error because, at that moment, your RAGFlow may not be fully initialized.
    >
 5. In your web browser, enter the IP address of your server and log in to RAGFlow.
@@ -302,6 +303,15 @@ cd ragflow/
 docker build --platform linux/amd64 -f Dockerfile -t infiniflow/ragflow:nightly .
 ```
 
+Or if you are behind a proxy, you can pass proxy arguments:
+
+```bash
+docker build --platform linux/amd64 \
+  --build-arg http_proxy=http://YOUR_PROXY:PORT \
+  --build-arg https_proxy=http://YOUR_PROXY:PORT \
+  -f Dockerfile -t infiniflow/ragflow:nightly .
+```
+
 ## 🔨 Launch service from source for development
 
 1. Install `uv` and `pre-commit`, or skip this step if they are already installed:

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:v0.22.1">
+        <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:v0.23.0">
     </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">
@@ -85,6 +85,7 @@ Coba demo kami di [https://demo.ragflow.io](https://demo.ragflow.io).
 
 ## 🔥 Pembaruan Terbaru
 
+- 2025-12-26 Mendukung 'Memori' untuk agen AI.
 - 2025-11-19 Mendukung Gemini 3 Pro.
 - 2025-11-12 Mendukung sinkronisasi data dari Confluence, S3, Notion, Discord, Google Drive.
 - 2025-10-23 Mendukung MinerU & Docling sebagai metode penguraian dokumen.
@@ -187,12 +188,12 @@ 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 v0.22.1 dari gambar Docker RAGFlow. Silakan merujuk ke tabel berikut untuk deskripsi berbagai edisi RAGFlow. Untuk mengunduh edisi RAGFlow yang berbeda dari v0.22.1, perbarui variabel RAGFLOW_IMAGE di docker/.env sebelum menggunakan docker compose untuk memulai server.
+> Perintah di bawah ini mengunduh edisi v0.23.0 dari gambar Docker RAGFlow. Silakan merujuk ke tabel berikut untuk deskripsi berbagai edisi RAGFlow. Untuk mengunduh edisi RAGFlow yang berbeda dari v0.23.0, perbarui variabel RAGFLOW_IMAGE di docker/.env sebelum menggunakan docker compose untuk memulai server.
 
 ```bash
    $ cd ragflow/docker
    
-   # git checkout v0.22.1
+   # git checkout v0.23.0
    # Opsional: gunakan tag stabil (lihat releases: https://github.com/infiniflow/ragflow/releases)
    # This steps ensures the **entrypoint.sh** file in the code matches the Docker image version.
 
@@ -232,7 +233,7 @@ Coba demo kami di [https://demo.ragflow.io](https://demo.ragflow.io).
     * Running on all addresses (0.0.0.0)
    ```
 
-   > Jika Anda melewatkan langkah ini dan langsung login ke RAGFlow, browser Anda mungkin menampilkan error `network anormal`
+   > Jika Anda melewatkan langkah ini dan langsung login ke RAGFlow, browser Anda mungkin menampilkan error `network abnormal`
    > karena RAGFlow mungkin belum sepenuhnya siap.
    >
 2. Buka browser web Anda, masukkan alamat IP server Anda, dan login ke RAGFlow.
@@ -276,6 +277,15 @@ cd ragflow/
 docker build --platform linux/amd64 -f Dockerfile -t infiniflow/ragflow:nightly .
 ```
 
+Jika berada di belakang proxy, Anda dapat melewatkan argumen proxy:
+
+```bash
+docker build --platform linux/amd64 \
+  --build-arg http_proxy=http://YOUR_PROXY:PORT \
+  --build-arg https_proxy=http://YOUR_PROXY:PORT \
+  -f Dockerfile -t infiniflow/ragflow:nightly .
+```
+
 ## 🔨 Menjalankan Aplikasi dari untuk Pengembangan
 
 1. Instal `uv` dan `pre-commit`, atau lewati langkah ini jika sudah terinstal:

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:v0.22.1">
+        <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:v0.23.0">
     </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">
@@ -66,7 +66,8 @@
 
 ## 🔥 最新情報
 
-- 2025-11-19 Gemini 3 Proをサポートしています
+- 2025-12-26 AIエージェントの「メモリ」機能をサポート。
+- 2025-11-19 Gemini 3 Proをサポートしています。
 - 2025-11-12 Confluence、S3、Notion、Discord、Google Drive からのデータ同期をサポートします。
 - 2025-10-23 ドキュメント解析方法として MinerU と Docling をサポートします。
 - 2025-10-15 オーケストレーションされたデータパイプラインのサポート。
@@ -167,12 +168,12 @@
 > 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 v0.22.1 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。v0.22.1 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。
+> 以下のコマンドは、RAGFlow Docker イメージの v0.23.0 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。v0.23.0 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。
 
 ```bash
    $ cd ragflow/docker
 
-   # git checkout v0.22.1
+   # git checkout v0.23.0
    # 任意: 安定版タグを利用 (一覧: https://github.com/infiniflow/ragflow/releases)
    # この手順は、コード内の entrypoint.sh ファイルが Docker イメージのバージョンと一致していることを確認します。
 
@@ -276,6 +277,15 @@ cd ragflow/
 docker build --platform linux/amd64 -f Dockerfile -t infiniflow/ragflow:nightly .
 ```
 
+プロキシ環境下にいる場合は、プロキシ引数を指定できます：
+
+```bash
+docker build --platform linux/amd64 \
+  --build-arg http_proxy=http://YOUR_PROXY:PORT \
+  --build-arg https_proxy=http://YOUR_PROXY:PORT \
+  -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` 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:

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:v0.22.1">
+        <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:v0.23.0">
     </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">
@@ -67,6 +67,7 @@
 
 ## 🔥 업데이트
 
+- 2025-12-26 AI 에이전트의 '메모리' 기능 지원.
 - 2025-11-19 Gemini 3 Pro를 지원합니다.
 - 2025-11-12 Confluence, S3, Notion, Discord, Google Drive에서 데이터 동기화를 지원합니다.
 - 2025-10-23 문서 파싱 방법으로 MinerU 및 Docling을 지원합니다.
@@ -169,12 +170,12 @@
 > 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 v0.22.1 버전을 다운로드합니다. 다양한 RAGFlow 버전에 대한 설명은 다음 표를 참조하십시오. v0.22.1과 다른 RAGFlow 버전을 다운로드하려면, docker/.env 파일에서 RAGFLOW_IMAGE 변수를 적절히 업데이트한 후 docker compose를 사용하여 서버를 시작하십시오.
+   > 아래 명령어는 RAGFlow Docker 이미지의 v0.23.0 버전을 다운로드합니다. 다양한 RAGFlow 버전에 대한 설명은 다음 표를 참조하십시오. v0.23.0과 다른 RAGFlow 버전을 다운로드하려면, docker/.env 파일에서 RAGFLOW_IMAGE 변수를 적절히 업데이트한 후 docker compose를 사용하여 서버를 시작하십시오.
 
    ```bash
    $ cd ragflow/docker
    
-   # git checkout v0.22.1
+   # git checkout v0.23.0
    # Optional: use a stable tag (see releases: https://github.com/infiniflow/ragflow/releases)
    # 이 단계는 코드의 entrypoint.sh 파일이 Docker 이미지 버전과 일치하도록 보장합니다.
 
@@ -213,7 +214,7 @@
     * Running on all addresses (0.0.0.0)
    ```
 
-   > 만약 확인 단계를 건너뛰고 바로 RAGFlow에 로그인하면, RAGFlow가 완전히 초기화되지 않았기 때문에 브라우저에서 `network anormal` 오류가 발생할 수 있습니다.
+   > 만약 확인 단계를 건너뛰고 바로 RAGFlow에 로그인하면, RAGFlow가 완전히 초기화되지 않았기 때문에 브라우저에서 `network abnormal` 오류가 발생할 수 있습니다.
 
 2. 웹 브라우저에 서버의 IP 주소를 입력하고 RAGFlow에 로그인하세요.
    > 기본 설정을 사용할 경우, `http://IP_OF_YOUR_MACHINE`만 입력하면 됩니다 (포트 번호는 제외). 기본 HTTP 서비스 포트 `80`은 기본 구성으로 사용할 때 생략할 수 있습니다.
@@ -270,6 +271,15 @@ cd ragflow/
 docker build --platform linux/amd64 -f Dockerfile -t infiniflow/ragflow:nightly .
 ```
 
+프록시 환경인 경우, 프록시 인수를 전달할 수 있습니다：
+
+```bash
+docker build --platform linux/amd64 \
+  --build-arg http_proxy=http://YOUR_PROXY:PORT \
+  --build-arg https_proxy=http://YOUR_PROXY:PORT \
+  -f Dockerfile -t infiniflow/ragflow:nightly .
+```
+
 ## 🔨 소스 코드로 서비스를 시작합니다.
 
 1. `uv` 와 `pre-commit` 을 설치하거나, 이미 설치된 경우 이 단계를 건너뜁니다:

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:v0.22.1">
+        <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:v0.23.0">
     </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">
@@ -86,6 +86,7 @@ Experimente nossa demo em [https://demo.ragflow.io](https://demo.ragflow.io).
 
 ## 🔥 Últimas Atualizações
 
+- 26-12-2025 Suporte à função 'Memória' para agentes de IA.
 - 19-11-2025 Suporta Gemini 3 Pro.
 - 12-11-2025 Suporta a sincronização de dados do Confluence, S3, Notion, Discord e Google Drive.
 - 23-10-2025 Suporta MinerU e Docling como métodos de análise de documentos.
@@ -187,12 +188,12 @@ 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`v0.22.1` 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 `v0.22.1`, atualize a variável `RAGFLOW_IMAGE` conforme necessário no **docker/.env** antes de usar `docker compose` para iniciar o servidor.
+    > O comando abaixo baixa a edição`v0.23.0` 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 `v0.23.0`, atualize a variável `RAGFLOW_IMAGE` conforme necessário no **docker/.env** antes de usar `docker compose` para iniciar o servidor.
 
 ```bash
    $ cd ragflow/docker
    
-   # git checkout v0.22.1
+   # git checkout v0.23.0
    # Opcional: use uma tag estável (veja releases: https://github.com/infiniflow/ragflow/releases)
    # Esta etapa garante que o arquivo entrypoint.sh no código corresponda à versão da imagem do Docker.
 
@@ -231,7 +232,7 @@ Experimente nossa demo em [https://demo.ragflow.io](https://demo.ragflow.io).
     * Rodando em todos os endereços (0.0.0.0)
    ```
 
-   > Se você pular essa etapa de confirmação e acessar diretamente o RAGFlow, seu navegador pode exibir um erro `network anormal`, pois, nesse momento, seu RAGFlow pode não estar totalmente inicializado.
+   > Se você pular essa etapa de confirmação e acessar diretamente o RAGFlow, seu navegador pode exibir um erro `network abnormal`, pois, nesse momento, seu RAGFlow pode não estar totalmente inicializado.
    >
 5. No seu navegador, insira o endereço IP do seu servidor e faça login no RAGFlow.
 
@@ -293,6 +294,15 @@ cd ragflow/
 docker build --platform linux/amd64 -f Dockerfile -t infiniflow/ragflow:nightly .
 ```
 
+Se você estiver atrás de um proxy, pode passar argumentos de proxy:
+
+```bash
+docker build --platform linux/amd64 \
+  --build-arg http_proxy=http://YOUR_PROXY:PORT \
+  --build-arg https_proxy=http://YOUR_PROXY:PORT \
+  -f Dockerfile -t infiniflow/ragflow:nightly .
+```
+
 ## 🔨 Lançar o serviço a partir do código-fonte para desenvolvimento
 
 1. Instale o `uv` e o `pre-commit`, ou pule esta etapa se eles já estiverem instalados:

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:v0.22.1">
+        <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:v0.23.0">
     </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">
@@ -85,15 +85,16 @@
 
 ## 🔥 近期更新
 
-- 2025-11-19 支援 Gemini 3 Pro.
+- 2025-12-26 支援AI代理的「記憶」功能。
+- 2025-11-19 支援 Gemini 3 Pro。
 - 2025-11-12 支援從 Confluence、S3、Notion、Discord、Google Drive 進行資料同步。
 - 2025-10-23 支援 MinerU 和 Docling 作為文件解析方法。
 - 2025-10-15 支援可編排的資料管道。
 - 2025-08-08 支援 OpenAI 最新的 GPT-5 系列模型。
-- 2025-08-01 支援 agentic workflow 和 MCP
+- 2025-08-01 支援 agentic workflow 和 MCP。
 - 2025-05-23 為 Agent 新增 Python/JS 程式碼執行器元件。
 - 2025-05-05 支援跨語言查詢。
-- 2025-03-19 PDF和DOCX中的圖支持用多模態大模型去解析得到描述.
+- 2025-03-19 PDF和DOCX中的圖支持用多模態大模型去解析得到描述。
 - 2024-12-18 升級了 DeepDoc 的文檔佈局分析模型。
 - 2024-08-22 支援用 RAG 技術實現從自然語言到 SQL 語句的轉換。
 
@@ -124,7 +125,7 @@
 
 ### 🍔 **相容各類異質資料來源**
 
-- 支援豐富的文件類型，包括 Word 文件、PPT、excel 表格、txt 檔案、圖片、PDF、影印件、影印件、結構化資料、網頁等。
+- 支援豐富的文件類型，包括 Word 文件、PPT、excel 表格、txt 檔案、圖片、PDF、影印件、複印件、結構化資料、網頁等。
 
 ### 🛀 **全程無憂、自動化的 RAG 工作流程**
 
@@ -186,12 +187,12 @@
 > 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 Docker 映像 `v0.22.1`。請參考下表查看不同 Docker 發行版的說明。如需下載不同於 `v0.22.1` 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。
+> 執行以下指令會自動下載 RAGFlow Docker 映像 `v0.23.0`。請參考下表查看不同 Docker 發行版的說明。如需下載不同於 `v0.23.0` 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。
 
 ```bash
    $ cd ragflow/docker
    
-   # git checkout v0.22.1
+   # git checkout v0.23.0
    # 可選：使用穩定版標籤（查看發佈：https://github.com/infiniflow/ragflow/releases）
    # 此步驟確保程式碼中的 entrypoint.sh 檔案與 Docker 映像版本一致。
 
@@ -236,7 +237,7 @@
     * Running on all addresses (0.0.0.0)
    ```
 
-   > 如果您跳過這一步驟系統確認步驟就登入 RAGFlow，你的瀏覽器有可能會提示 `network anormal` 或 `網路異常`，因為 RAGFlow 可能並未完全啟動成功。
+   > 如果您跳過這一步驟系統確認步驟就登入 RAGFlow，你的瀏覽器有可能會提示 `network abnormal` 或 `網路異常`，因為 RAGFlow 可能並未完全啟動成功。
    >
 5. 在你的瀏覽器中輸入你的伺服器對應的 IP 位址並登入 RAGFlow。
 
@@ -302,6 +303,15 @@ cd ragflow/
 docker build --platform linux/amd64 -f Dockerfile -t infiniflow/ragflow:nightly .
 ```
 
+若您位於代理環境，可傳遞代理參數：
+
+```bash
+docker build --platform linux/amd64 \
+  --build-arg http_proxy=http://YOUR_PROXY:PORT \
+  --build-arg https_proxy=http://YOUR_PROXY:PORT \
+  -f Dockerfile -t infiniflow/ragflow:nightly .
+```
+
 ## 🔨 以原始碼啟動服務
 
 1. 安裝 `uv` 和 `pre-commit`。如已安裝，可跳過此步驟：

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:v0.22.1">
+        <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:v0.23.0">
     </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">
@@ -85,7 +85,8 @@
 
 ## 🔥 近期更新
 
-- 2025-11-19 支持 Gemini 3 Pro.
+- 2025-12-26 支持AI代理的“记忆”功能。
+- 2025-11-19 支持 Gemini 3 Pro。
 - 2025-11-12 支持从 Confluence、S3、Notion、Discord、Google Drive 进行数据同步。
 - 2025-10-23 支持 MinerU 和 Docling 作为文档解析方法。
 - 2025-10-15 支持可编排的数据管道。
@@ -93,7 +94,7 @@
 - 2025-08-01 支持 agentic workflow 和 MCP。
 - 2025-05-23 Agent 新增 Python/JS 代码执行器组件。
 - 2025-05-05 支持跨语言查询。
-- 2025-03-19 PDF 和 DOCX 中的图支持用多模态大模型去解析得到描述.
+- 2025-03-19 PDF 和 DOCX 中的图支持用多模态大模型去解析得到描述。
 - 2024-12-18 升级了 DeepDoc 的文档布局分析模型。
 - 2024-08-22 支持用 RAG 技术实现从自然语言到 SQL 语句的转换。
 
@@ -187,12 +188,12 @@
 > 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 Docker 镜像 `v0.22.1`。请参考下表查看不同 Docker 发行版的描述。如需下载不同于 `v0.22.1` 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。
+   > 运行以下命令会自动下载 RAGFlow Docker 镜像 `v0.23.0`。请参考下表查看不同 Docker 发行版的描述。如需下载不同于 `v0.23.0` 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。
 
    ```bash
    $ cd ragflow/docker
    
-   # git checkout v0.22.1
+   # git checkout v0.23.0
    # 可选：使用稳定版本标签（查看发布：https://github.com/infiniflow/ragflow/releases）
    # 这一步确保代码中的 entrypoint.sh 文件与 Docker 镜像的版本保持一致。
 
@@ -237,7 +238,7 @@
     * Running on all addresses (0.0.0.0)
    ```
 
-   > 如果您在没有看到上面的提示信息出来之前，就尝试登录 RAGFlow，你的浏览器有可能会提示 `network anormal` 或 `网络异常`。
+   > 如果您在没有看到上面的提示信息出来之前，就尝试登录 RAGFlow，你的浏览器有可能会提示 `network abnormal` 或 `网络异常`。
 
 5. 在你的浏览器中输入你的服务器对应的 IP 地址并登录 RAGFlow。
    > 上面这个例子中，您只需输入 http://IP_OF_YOUR_MACHINE 即可：未改动过配置则无需输入端口（默认的 HTTP 服务端口 80）。
@@ -301,6 +302,15 @@ cd ragflow/
 docker build --platform linux/amd64 -f Dockerfile -t infiniflow/ragflow:nightly .
 ```
 
+如果您处在代理环境下，可以传递代理参数：
+
+```bash
+docker build --platform linux/amd64 \
+  --build-arg http_proxy=http://YOUR_PROXY:PORT \
+  --build-arg https_proxy=http://YOUR_PROXY:PORT \
+  -f Dockerfile -t infiniflow/ragflow:nightly .
+```
+
 ## 🔨 以源代码启动服务
 
 1. 安装 `uv` 和 `pre-commit`。如已经安装，可跳过本步骤：

admin/client/README.md

@@ -48,7 +48,7 @@ It consists of a server-side Service and a command-line client (CLI), both imple
 1.  Ensure the Admin Service is running.
 2.  Install ragflow-cli.
     ```bash
-    pip install ragflow-cli==0.22.1
+    pip install ragflow-cli==0.23.0
     ```
 3.  Launch the CLI client:
     ```bash

admin/client/admin_client.py

@@ -16,14 +16,14 @@
 
 import argparse
 import base64
-from cmd import Cmd
-
-from Cryptodome.PublicKey import RSA
-from Cryptodome.Cipher import PKCS1_v1_5 as Cipher_pkcs1_v1_5
-from typing import Dict, List, Any
-from lark import Lark, Transformer, Tree
-import requests
 import getpass
+from cmd import Cmd
+from typing import Any, Dict, List
+
+import requests
+from Cryptodome.Cipher import PKCS1_v1_5 as Cipher_pkcs1_v1_5
+from Cryptodome.PublicKey import RSA
+from lark import Lark, Transformer, Tree
 
 GRAMMAR = r"""
 start: command
@@ -141,7 +141,6 @@ NUMBER: /[0-9]+/
 
 
 class AdminTransformer(Transformer):
-
     def start(self, items):
         return items[0]
 
@@ -149,7 +148,7 @@ class AdminTransformer(Transformer):
         return items[0]
 
     def list_services(self, items):
-        result = {'type': 'list_services'}
+        result = {"type": "list_services"}
         return result
 
     def show_service(self, items):
@@ -236,11 +235,7 @@ class AdminTransformer(Transformer):
         action_list = items[1]
         resource = items[3]
         role_name = items[6]
-        return {
-            "type": "revoke_permission",
-            "role_name": role_name,
-            "resource": resource, "actions": action_list
-        }
+        return {"type": "revoke_permission", "role_name": role_name, "resource": resource, "actions": action_list}
 
     def alter_user_role(self, items):
         user_name = items[2]
@@ -264,12 +259,12 @@ class AdminTransformer(Transformer):
         # handle quoted parameter
         parsed_args = []
         for arg in args:
-            if hasattr(arg, 'value'):
+            if hasattr(arg, "value"):
                 parsed_args.append(arg.value)
             else:
                 parsed_args.append(str(arg))
 
-        return {'type': 'meta', 'command': command_name, 'args': parsed_args}
+        return {"type": "meta", "command": command_name, "args": parsed_args}
 
     def meta_command_name(self, items):
         return items[0]
@@ -279,22 +274,22 @@ class AdminTransformer(Transformer):
 
 
 def encrypt(input_string):
-    pub = '-----BEGIN PUBLIC KEY-----\nMIIBIjANBgkqhkiG9w0BAQEFAAOCAQ8AMIIBCgKCAQEArq9XTUSeYr2+N1h3Afl/z8Dse/2yD0ZGrKwx+EEEcdsBLca9Ynmx3nIB5obmLlSfmskLpBo0UACBmB5rEjBp2Q2f3AG3Hjd4B+gNCG6BDaawuDlgANIhGnaTLrIqWrrcm4EMzJOnAOI1fgzJRsOOUEfaS318Eq9OVO3apEyCCt0lOQK6PuksduOjVxtltDav+guVAA068NrPYmRNabVKRNLJpL8w4D44sfth5RvZ3q9t+6RTArpEtc5sh5ChzvqPOzKGMXW83C95TxmXqpbK6olN4RevSfVjEAgCydH6HN6OhtOQEcnrU97r9H0iZOWwbw3pVrZiUkuRD1R56Wzs2wIDAQAB\n-----END PUBLIC KEY-----'
+    pub = "-----BEGIN PUBLIC KEY-----\nMIIBIjANBgkqhkiG9w0BAQEFAAOCAQ8AMIIBCgKCAQEArq9XTUSeYr2+N1h3Afl/z8Dse/2yD0ZGrKwx+EEEcdsBLca9Ynmx3nIB5obmLlSfmskLpBo0UACBmB5rEjBp2Q2f3AG3Hjd4B+gNCG6BDaawuDlgANIhGnaTLrIqWrrcm4EMzJOnAOI1fgzJRsOOUEfaS318Eq9OVO3apEyCCt0lOQK6PuksduOjVxtltDav+guVAA068NrPYmRNabVKRNLJpL8w4D44sfth5RvZ3q9t+6RTArpEtc5sh5ChzvqPOzKGMXW83C95TxmXqpbK6olN4RevSfVjEAgCydH6HN6OhtOQEcnrU97r9H0iZOWwbw3pVrZiUkuRD1R56Wzs2wIDAQAB\n-----END PUBLIC KEY-----"
     pub_key = RSA.importKey(pub)
     cipher = Cipher_pkcs1_v1_5.new(pub_key)
-    cipher_text = cipher.encrypt(base64.b64encode(input_string.encode('utf-8')))
+    cipher_text = cipher.encrypt(base64.b64encode(input_string.encode("utf-8")))
     return base64.b64encode(cipher_text).decode("utf-8")
 
 
 def encode_to_base64(input_string):
-    base64_encoded = base64.b64encode(input_string.encode('utf-8'))
-    return base64_encoded.decode('utf-8')
+    base64_encoded = base64.b64encode(input_string.encode("utf-8"))
+    return base64_encoded.decode("utf-8")
 
 
 class AdminCLI(Cmd):
     def __init__(self):
         super().__init__()
-        self.parser = Lark(GRAMMAR, start='start', parser='lalr', transformer=AdminTransformer())
+        self.parser = Lark(GRAMMAR, start="start", parser="lalr", transformer=AdminTransformer())
         self.command_history = []
         self.is_interactive = False
         self.admin_account = "admin@ragflow.io"
@@ -312,7 +307,7 @@ class AdminCLI(Cmd):
             result = self.parse_command(command)
 
             if isinstance(result, dict):
-                if 'type' in result and result.get('type') == 'empty':
+                if "type" in result and result.get("type") == "empty":
                     return False
 
             self.execute_command(result)
@@ -320,7 +315,7 @@ class AdminCLI(Cmd):
             if isinstance(result, Tree):
                 return False
 
-            if result.get('type') == 'meta' and result.get('command') in ['q', 'quit', 'exit']:
+            if result.get("type") == "meta" and result.get("command") in ["q", "quit", "exit"]:
                 return True
 
         except KeyboardInterrupt:
@@ -338,7 +333,7 @@ class AdminCLI(Cmd):
 
     def parse_command(self, command_str: str) -> dict[str, str]:
         if not command_str.strip():
-            return {'type': 'empty'}
+            return {"type": "empty"}
 
         self.command_history.append(command_str)
 
@@ -346,11 +341,11 @@ class AdminCLI(Cmd):
             result = self.parser.parse(command_str)
             return result
         except Exception as e:
-            return {'type': 'error', 'message': f'Parse error: {str(e)}'}
+            return {"type": "error", "message": f"Parse error: {str(e)}"}
 
     def verify_admin(self, arguments: dict, single_command: bool):
-        self.host = arguments['host']
-        self.port = arguments['port']
+        self.host = arguments["host"]
+        self.port = arguments["port"]
         print("Attempt to access server for admin login")
         url = f"http://{self.host}:{self.port}/api/v1/admin/login"
 
@@ -365,25 +360,21 @@ class AdminCLI(Cmd):
                 return False
 
             if single_command:
-                admin_passwd = arguments['password']
+                admin_passwd = arguments["password"]
             else:
                 admin_passwd = getpass.getpass(f"password for {self.admin_account}: ").strip()
             try:
                 self.admin_password = encrypt(admin_passwd)
-                response = self.session.post(url, json={'email': self.admin_account, 'password': self.admin_password})
+                response = self.session.post(url, json={"email": self.admin_account, "password": self.admin_password})
                 if response.status_code == 200:
                     res_json = response.json()
-                    error_code = res_json.get('code', -1)
+                    error_code = res_json.get("code", -1)
                     if error_code == 0:
-                        self.session.headers.update({
-                            'Content-Type': 'application/json',
-                            'Authorization': response.headers['Authorization'],
-                            'User-Agent': 'RAGFlow-CLI/0.22.1'
-                        })
+                        self.session.headers.update({"Content-Type": "application/json", "Authorization": response.headers["Authorization"], "User-Agent": "RAGFlow-CLI/0.23.0"})
                         print("Authentication successful.")
                         return True
                     else:
-                        error_message = res_json.get('message', 'Unknown error')
+                        error_message = res_json.get("message", "Unknown error")
                         print(f"Authentication failed: {error_message}, try again")
                         continue
                 else:
@@ -403,10 +394,14 @@ class AdminCLI(Cmd):
         for k, v in data.items():
             # display latest status
             heartbeats = sorted(v, key=lambda x: x["now"], reverse=True)
-            task_executor_list.append({
-                "task_executor_name": k,
-                **heartbeats[0],
-            } if heartbeats else {"task_executor_name": k})
+            task_executor_list.append(
+                {
+                    "task_executor_name": k,
+                    **heartbeats[0],
+                }
+                if heartbeats
+                else {"task_executor_name": k}
+            )
         return task_executor_list
 
     def _print_table_simple(self, data):
@@ -422,12 +417,7 @@ class AdminCLI(Cmd):
         col_widths = {}
 
         def get_string_width(text):
-            half_width_chars = (
-                " !\"#$%&'()*+,-./0123456789:;<=>?@"
-                "ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_`"
-                "abcdefghijklmnopqrstuvwxyz{|}~"
-                "\t\n\r"
-            )
+            half_width_chars = " !\"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_`abcdefghijklmnopqrstuvwxyz{|}~\t\n\r"
             width = 0
             for char in text:
                 if char in half_width_chars:
@@ -439,7 +429,7 @@ class AdminCLI(Cmd):
         for col in columns:
             max_width = get_string_width(str(col))
             for item in data:
-                value_len = get_string_width(str(item.get(col, '')))
+                value_len = get_string_width(str(item.get(col, "")))
                 if value_len > max_width:
                     max_width = value_len
             col_widths[col] = max(2, max_width)
@@ -457,16 +447,15 @@ class AdminCLI(Cmd):
         for item in data:
             row = "|"
             for col in columns:
-                value = str(item.get(col, ''))
+                value = str(item.get(col, ""))
                 if get_string_width(value) > col_widths[col]:
-                    value = value[:col_widths[col] - 3] + "..."
+                    value = value[: col_widths[col] - 3] + "..."
                 row += f" {value:<{col_widths[col] - (get_string_width(value) - len(value))}} |"
             print(row)
 
         print(separator)
 
     def run_interactive(self):
-
         self.is_interactive = True
         print("RAGFlow Admin command line interface - Type '\\?' for help, '\\q' to quit")
 
@@ -483,7 +472,7 @@ class AdminCLI(Cmd):
                 if isinstance(result, Tree):
                     continue
 
-                if result.get('type') == 'meta' and result.get('command') in ['q', 'quit', 'exit']:
+                if result.get("type") == "meta" and result.get("command") in ["q", "quit", "exit"]:
                     break
 
             except KeyboardInterrupt:
@@ -497,36 +486,30 @@ class AdminCLI(Cmd):
         self.execute_command(result)
 
     def parse_connection_args(self, args: List[str]) -> Dict[str, Any]:
-        parser = argparse.ArgumentParser(description='Admin CLI Client', add_help=False)
-        parser.add_argument('-h', '--host', default='localhost', help='Admin service host')
-        parser.add_argument('-p', '--port', type=int, default=9381, help='Admin service port')
-        parser.add_argument('-w', '--password', default='admin', type=str, help='Superuser password')
-        parser.add_argument('command', nargs='?', help='Single command')
+        parser = argparse.ArgumentParser(description="Admin CLI Client", add_help=False)
+        parser.add_argument("-h", "--host", default="localhost", help="Admin service host")
+        parser.add_argument("-p", "--port", type=int, default=9381, help="Admin service port")
+        parser.add_argument("-w", "--password", default="admin", type=str, help="Superuser password")
+        parser.add_argument("command", nargs="?", help="Single command")
         try:
             parsed_args, remaining_args = parser.parse_known_args(args)
             if remaining_args:
                 command = remaining_args[0]
-                return {
-                    'host': parsed_args.host,
-                    'port': parsed_args.port,
-                    'password': parsed_args.password,
-                    'command': command
-                }
+                return {"host": parsed_args.host, "port": parsed_args.port, "password": parsed_args.password, "command": command}
             else:
                 return {
-                    'host': parsed_args.host,
-                    'port': parsed_args.port,
+                    "host": parsed_args.host,
+                    "port": parsed_args.port,
                 }
         except SystemExit:
-            return {'error': 'Invalid connection arguments'}
+            return {"error": "Invalid connection arguments"}
 
     def execute_command(self, parsed_command: Dict[str, Any]):
-
         command_dict: dict
         if isinstance(parsed_command, Tree):
             command_dict = parsed_command.children[0]
         else:
-            if parsed_command['type'] == 'error':
+            if parsed_command["type"] == "error":
                 print(f"Error: {parsed_command['message']}")
                 return
             else:
@@ -534,56 +517,56 @@ class AdminCLI(Cmd):
 
         # print(f"Parsed command: {command_dict}")
 
-        command_type = command_dict['type']
+        command_type = command_dict["type"]
 
         match command_type:
-            case 'list_services':
+            case "list_services":
                 self._handle_list_services(command_dict)
-            case 'show_service':
+            case "show_service":
                 self._handle_show_service(command_dict)
-            case 'restart_service':
+            case "restart_service":
                 self._handle_restart_service(command_dict)
-            case 'shutdown_service':
+            case "shutdown_service":
                 self._handle_shutdown_service(command_dict)
-            case 'startup_service':
+            case "startup_service":
                 self._handle_startup_service(command_dict)
-            case 'list_users':
+            case "list_users":
                 self._handle_list_users(command_dict)
-            case 'show_user':
+            case "show_user":
                 self._handle_show_user(command_dict)
-            case 'drop_user':
+            case "drop_user":
                 self._handle_drop_user(command_dict)
-            case 'alter_user':
+            case "alter_user":
                 self._handle_alter_user(command_dict)
-            case 'create_user':
+            case "create_user":
                 self._handle_create_user(command_dict)
-            case 'activate_user':
+            case "activate_user":
                 self._handle_activate_user(command_dict)
-            case 'list_datasets':
+            case "list_datasets":
                 self._handle_list_datasets(command_dict)
-            case 'list_agents':
+            case "list_agents":
                 self._handle_list_agents(command_dict)
-            case 'create_role':
+            case "create_role":
                 self._create_role(command_dict)
-            case 'drop_role':
+            case "drop_role":
                 self._drop_role(command_dict)
-            case 'alter_role':
+            case "alter_role":
                 self._alter_role(command_dict)
-            case 'list_roles':
+            case "list_roles":
                 self._list_roles(command_dict)
-            case 'show_role':
+            case "show_role":
                 self._show_role(command_dict)
-            case 'grant_permission':
+            case "grant_permission":
                 self._grant_permission(command_dict)
-            case 'revoke_permission':
+            case "revoke_permission":
                 self._revoke_permission(command_dict)
-            case 'alter_user_role':
+            case "alter_user_role":
                 self._alter_user_role(command_dict)
-            case 'show_user_permission':
+            case "show_user_permission":
                 self._show_user_permission(command_dict)
-            case 'show_version':
+            case "show_version":
                 self._show_version(command_dict)
-            case 'meta':
+            case "meta":
                 self._handle_meta_command(command_dict)
             case _:
                 print(f"Command '{command_type}' would be executed with API")
@@ -591,29 +574,29 @@ class AdminCLI(Cmd):
     def _handle_list_services(self, command):
         print("Listing all services")
 
-        url = f'http://{self.host}:{self.port}/api/v1/admin/services'
+        url = f"http://{self.host}:{self.port}/api/v1/admin/services"
         response = self.session.get(url)
         res_json = response.json()
         if response.status_code == 200:
-            self._print_table_simple(res_json['data'])
+            self._print_table_simple(res_json["data"])
         else:
             print(f"Fail to get all services, code: {res_json['code']}, message: {res_json['message']}")
 
     def _handle_show_service(self, command):
-        service_id: int = command['number']
+        service_id: int = command["number"]
         print(f"Showing service: {service_id}")
 
-        url = f'http://{self.host}:{self.port}/api/v1/admin/services/{service_id}'
+        url = f"http://{self.host}:{self.port}/api/v1/admin/services/{service_id}"
         response = self.session.get(url)
         res_json = response.json()
         if response.status_code == 200:
-            res_data = res_json['data']
-            if 'status' in res_data and res_data['status'] == 'alive':
+            res_data = res_json["data"]
+            if "status" in res_data and res_data["status"] == "alive":
                 print(f"Service {res_data['service_name']} is alive, ")
-                if isinstance(res_data['message'], str):
-                    print(res_data['message'])
+                if isinstance(res_data["message"], str):
+                    print(res_data["message"])
                 else:
-                    data = self._format_service_detail_table(res_data['message'])
+                    data = self._format_service_detail_table(res_data["message"])
                     self._print_table_simple(data)
             else:
                 print(f"Service {res_data['service_name']} is down, {res_data['message']}")
@@ -621,47 +604,47 @@ class AdminCLI(Cmd):
             print(f"Fail to show service, code: {res_json['code']}, message: {res_json['message']}")
 
     def _handle_restart_service(self, command):
-        service_id: int = command['number']
+        service_id: int = command["number"]
         print(f"Restart service {service_id}")
 
     def _handle_shutdown_service(self, command):
-        service_id: int = command['number']
+        service_id: int = command["number"]
         print(f"Shutdown service {service_id}")
 
     def _handle_startup_service(self, command):
-        service_id: int = command['number']
+        service_id: int = command["number"]
         print(f"Startup service {service_id}")
 
     def _handle_list_users(self, command):
         print("Listing all users")
 
-        url = f'http://{self.host}:{self.port}/api/v1/admin/users'
+        url = f"http://{self.host}:{self.port}/api/v1/admin/users"
         response = self.session.get(url)
         res_json = response.json()
         if response.status_code == 200:
-            self._print_table_simple(res_json['data'])
+            self._print_table_simple(res_json["data"])
         else:
             print(f"Fail to get all users, code: {res_json['code']}, message: {res_json['message']}")
 
     def _handle_show_user(self, command):
-        username_tree: Tree = command['user_name']
+        username_tree: Tree = command["user_name"]
         user_name: str = username_tree.children[0].strip("'\"")
         print(f"Showing user: {user_name}")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/users/{user_name}'
+        url = f"http://{self.host}:{self.port}/api/v1/admin/users/{user_name}"
         response = self.session.get(url)
         res_json = response.json()
         if response.status_code == 200:
-            table_data = res_json['data']
-            table_data.pop('avatar')
+            table_data = res_json["data"]
+            table_data.pop("avatar")
             self._print_table_simple(table_data)
         else:
             print(f"Fail to get user {user_name}, code: {res_json['code']}, message: {res_json['message']}")
 
     def _handle_drop_user(self, command):
-        username_tree: Tree = command['user_name']
+        username_tree: Tree = command["user_name"]
         user_name: str = username_tree.children[0].strip("'\"")
         print(f"Drop user: {user_name}")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/users/{user_name}'
+        url = f"http://{self.host}:{self.port}/api/v1/admin/users/{user_name}"
         response = self.session.delete(url)
         res_json = response.json()
         if response.status_code == 200:
@@ -670,13 +653,13 @@ class AdminCLI(Cmd):
             print(f"Fail to drop user, code: {res_json['code']}, message: {res_json['message']}")
 
     def _handle_alter_user(self, command):
-        user_name_tree: Tree = command['user_name']
+        user_name_tree: Tree = command["user_name"]
         user_name: str = user_name_tree.children[0].strip("'\"")
-        password_tree: Tree = command['password']
+        password_tree: Tree = command["password"]
         password: str = password_tree.children[0].strip("'\"")
         print(f"Alter user: {user_name}, password: ******")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/users/{user_name}/password'
-        response = self.session.put(url, json={'new_password': encrypt(password)})
+        url = f"http://{self.host}:{self.port}/api/v1/admin/users/{user_name}/password"
+        response = self.session.put(url, json={"new_password": encrypt(password)})
         res_json = response.json()
         if response.status_code == 200:
             print(res_json["message"])
@@ -684,32 +667,29 @@ class AdminCLI(Cmd):
             print(f"Fail to alter password, code: {res_json['code']}, message: {res_json['message']}")
 
     def _handle_create_user(self, command):
-        user_name_tree: Tree = command['user_name']
+        user_name_tree: Tree = command["user_name"]
         user_name: str = user_name_tree.children[0].strip("'\"")
-        password_tree: Tree = command['password']
+        password_tree: Tree = command["password"]
         password: str = password_tree.children[0].strip("'\"")
-        role: str = command['role']
+        role: str = command["role"]
         print(f"Create user: {user_name}, password: ******, role: {role}")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/users'
-        response = self.session.post(
-            url,
-            json={'user_name': user_name, 'password': encrypt(password), 'role': role}
-        )
+        url = f"http://{self.host}:{self.port}/api/v1/admin/users"
+        response = self.session.post(url, json={"user_name": user_name, "password": encrypt(password), "role": role})
         res_json = response.json()
         if response.status_code == 200:
-            self._print_table_simple(res_json['data'])
+            self._print_table_simple(res_json["data"])
         else:
             print(f"Fail to create user {user_name}, code: {res_json['code']}, message: {res_json['message']}")
 
     def _handle_activate_user(self, command):
-        user_name_tree: Tree = command['user_name']
+        user_name_tree: Tree = command["user_name"]
         user_name: str = user_name_tree.children[0].strip("'\"")
-        activate_tree: Tree = command['activate_status']
+        activate_tree: Tree = command["activate_status"]
         activate_status: str = activate_tree.children[0].strip("'\"")
-        if activate_status.lower() in ['on', 'off']:
+        if activate_status.lower() in ["on", "off"]:
             print(f"Alter user {user_name} activate status, turn {activate_status.lower()}.")
-            url = f'http://{self.host}:{self.port}/api/v1/admin/users/{user_name}/activate'
-            response = self.session.put(url, json={'activate_status': activate_status})
+            url = f"http://{self.host}:{self.port}/api/v1/admin/users/{user_name}/activate"
+            response = self.session.put(url, json={"activate_status": activate_status})
             res_json = response.json()
             if response.status_code == 200:
                 print(res_json["message"])
@@ -719,202 +699,182 @@ class AdminCLI(Cmd):
             print(f"Unknown activate status: {activate_status}.")
 
     def _handle_list_datasets(self, command):
-        username_tree: Tree = command['user_name']
+        username_tree: Tree = command["user_name"]
         user_name: str = username_tree.children[0].strip("'\"")
         print(f"Listing all datasets of user: {user_name}")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/users/{user_name}/datasets'
+        url = f"http://{self.host}:{self.port}/api/v1/admin/users/{user_name}/datasets"
         response = self.session.get(url)
         res_json = response.json()
         if response.status_code == 200:
-            table_data = res_json['data']
+            table_data = res_json["data"]
             for t in table_data:
-                t.pop('avatar')
+                t.pop("avatar")
             self._print_table_simple(table_data)
         else:
             print(f"Fail to get all datasets of {user_name}, code: {res_json['code']}, message: {res_json['message']}")
 
     def _handle_list_agents(self, command):
-        username_tree: Tree = command['user_name']
+        username_tree: Tree = command["user_name"]
         user_name: str = username_tree.children[0].strip("'\"")
         print(f"Listing all agents of user: {user_name}")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/users/{user_name}/agents'
+        url = f"http://{self.host}:{self.port}/api/v1/admin/users/{user_name}/agents"
         response = self.session.get(url)
         res_json = response.json()
         if response.status_code == 200:
-            table_data = res_json['data']
+            table_data = res_json["data"]
             for t in table_data:
-                t.pop('avatar')
+                t.pop("avatar")
             self._print_table_simple(table_data)
         else:
             print(f"Fail to get all agents of {user_name}, code: {res_json['code']}, message: {res_json['message']}")
 
     def _create_role(self, command):
-        role_name_tree: Tree = command['role_name']
+        role_name_tree: Tree = command["role_name"]
         role_name: str = role_name_tree.children[0].strip("'\"")
-        desc_str: str = ''
-        if 'description' in command:
-            desc_tree: Tree = command['description']
+        desc_str: str = ""
+        if "description" in command:
+            desc_tree: Tree = command["description"]
             desc_str = desc_tree.children[0].strip("'\"")
 
         print(f"create role name: {role_name}, description: {desc_str}")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/roles'
-        response = self.session.post(
-            url,
-            json={'role_name': role_name, 'description': desc_str}
-        )
+        url = f"http://{self.host}:{self.port}/api/v1/admin/roles"
+        response = self.session.post(url, json={"role_name": role_name, "description": desc_str})
         res_json = response.json()
         if response.status_code == 200:
-            self._print_table_simple(res_json['data'])
+            self._print_table_simple(res_json["data"])
         else:
             print(f"Fail to create role {role_name}, code: {res_json['code']}, message: {res_json['message']}")
 
     def _drop_role(self, command):
-        role_name_tree: Tree = command['role_name']
+        role_name_tree: Tree = command["role_name"]
         role_name: str = role_name_tree.children[0].strip("'\"")
         print(f"drop role name: {role_name}")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/roles/{role_name}'
+        url = f"http://{self.host}:{self.port}/api/v1/admin/roles/{role_name}"
         response = self.session.delete(url)
         res_json = response.json()
         if response.status_code == 200:
-            self._print_table_simple(res_json['data'])
+            self._print_table_simple(res_json["data"])
         else:
             print(f"Fail to drop role {role_name}, code: {res_json['code']}, message: {res_json['message']}")
 
     def _alter_role(self, command):
-        role_name_tree: Tree = command['role_name']
+        role_name_tree: Tree = command["role_name"]
         role_name: str = role_name_tree.children[0].strip("'\"")
-        desc_tree: Tree = command['description']
+        desc_tree: Tree = command["description"]
         desc_str: str = desc_tree.children[0].strip("'\"")
 
         print(f"alter role name: {role_name}, description: {desc_str}")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/roles/{role_name}'
-        response = self.session.put(
-            url,
-            json={'description': desc_str}
-        )
+        url = f"http://{self.host}:{self.port}/api/v1/admin/roles/{role_name}"
+        response = self.session.put(url, json={"description": desc_str})
         res_json = response.json()
         if response.status_code == 200:
-            self._print_table_simple(res_json['data'])
+            self._print_table_simple(res_json["data"])
         else:
-            print(
-                f"Fail to update role {role_name} with description: {desc_str}, code: {res_json['code']}, message: {res_json['message']}")
+            print(f"Fail to update role {role_name} with description: {desc_str}, code: {res_json['code']}, message: {res_json['message']}")
 
     def _list_roles(self, command):
         print("Listing all roles")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/roles'
+        url = f"http://{self.host}:{self.port}/api/v1/admin/roles"
         response = self.session.get(url)
         res_json = response.json()
         if response.status_code == 200:
-            self._print_table_simple(res_json['data'])
+            self._print_table_simple(res_json["data"])
         else:
             print(f"Fail to list roles, code: {res_json['code']}, message: {res_json['message']}")
 
     def _show_role(self, command):
-        role_name_tree: Tree = command['role_name']
+        role_name_tree: Tree = command["role_name"]
         role_name: str = role_name_tree.children[0].strip("'\"")
         print(f"show role: {role_name}")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/roles/{role_name}/permission'
+        url = f"http://{self.host}:{self.port}/api/v1/admin/roles/{role_name}/permission"
         response = self.session.get(url)
         res_json = response.json()
         if response.status_code == 200:
-            self._print_table_simple(res_json['data'])
+            self._print_table_simple(res_json["data"])
         else:
             print(f"Fail to list roles, code: {res_json['code']}, message: {res_json['message']}")
 
     def _grant_permission(self, command):
-        role_name_tree: Tree = command['role_name']
+        role_name_tree: Tree = command["role_name"]
         role_name_str: str = role_name_tree.children[0].strip("'\"")
-        resource_tree: Tree = command['resource']
+        resource_tree: Tree = command["resource"]
         resource_str: str = resource_tree.children[0].strip("'\"")
-        action_tree_list: list = command['actions']
+        action_tree_list: list = command["actions"]
         actions: list = []
         for action_tree in action_tree_list:
             action_str: str = action_tree.children[0].strip("'\"")
             actions.append(action_str)
         print(f"grant role_name: {role_name_str}, resource: {resource_str}, actions: {actions}")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/roles/{role_name_str}/permission'
-        response = self.session.post(
-            url,
-            json={'actions': actions, 'resource': resource_str}
-        )
+        url = f"http://{self.host}:{self.port}/api/v1/admin/roles/{role_name_str}/permission"
+        response = self.session.post(url, json={"actions": actions, "resource": resource_str})
         res_json = response.json()
         if response.status_code == 200:
-            self._print_table_simple(res_json['data'])
+            self._print_table_simple(res_json["data"])
         else:
-            print(
-                f"Fail to grant role {role_name_str} with {actions} on {resource_str}, code: {res_json['code']}, message: {res_json['message']}")
+            print(f"Fail to grant role {role_name_str} with {actions} on {resource_str}, code: {res_json['code']}, message: {res_json['message']}")
 
     def _revoke_permission(self, command):
-        role_name_tree: Tree = command['role_name']
+        role_name_tree: Tree = command["role_name"]
         role_name_str: str = role_name_tree.children[0].strip("'\"")
-        resource_tree: Tree = command['resource']
+        resource_tree: Tree = command["resource"]
         resource_str: str = resource_tree.children[0].strip("'\"")
-        action_tree_list: list = command['actions']
+        action_tree_list: list = command["actions"]
         actions: list = []
         for action_tree in action_tree_list:
             action_str: str = action_tree.children[0].strip("'\"")
             actions.append(action_str)
         print(f"revoke role_name: {role_name_str}, resource: {resource_str}, actions: {actions}")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/roles/{role_name_str}/permission'
-        response = self.session.delete(
-            url,
-            json={'actions': actions, 'resource': resource_str}
-        )
+        url = f"http://{self.host}:{self.port}/api/v1/admin/roles/{role_name_str}/permission"
+        response = self.session.delete(url, json={"actions": actions, "resource": resource_str})
         res_json = response.json()
         if response.status_code == 200:
-            self._print_table_simple(res_json['data'])
+            self._print_table_simple(res_json["data"])
         else:
-            print(
-                f"Fail to revoke role {role_name_str} with {actions} on {resource_str}, code: {res_json['code']}, message: {res_json['message']}")
+            print(f"Fail to revoke role {role_name_str} with {actions} on {resource_str}, code: {res_json['code']}, message: {res_json['message']}")
 
     def _alter_user_role(self, command):
-        role_name_tree: Tree = command['role_name']
+        role_name_tree: Tree = command["role_name"]
         role_name_str: str = role_name_tree.children[0].strip("'\"")
-        user_name_tree: Tree = command['user_name']
+        user_name_tree: Tree = command["user_name"]
         user_name_str: str = user_name_tree.children[0].strip("'\"")
         print(f"alter_user_role user_name: {user_name_str}, role_name: {role_name_str}")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/users/{user_name_str}/role'
-        response = self.session.put(
-            url,
-            json={'role_name': role_name_str}
-        )
+        url = f"http://{self.host}:{self.port}/api/v1/admin/users/{user_name_str}/role"
+        response = self.session.put(url, json={"role_name": role_name_str})
         res_json = response.json()
         if response.status_code == 200:
-            self._print_table_simple(res_json['data'])
+            self._print_table_simple(res_json["data"])
         else:
-            print(
-                f"Fail to alter user: {user_name_str} to role {role_name_str}, code: {res_json['code']}, message: {res_json['message']}")
+            print(f"Fail to alter user: {user_name_str} to role {role_name_str}, code: {res_json['code']}, message: {res_json['message']}")
 
     def _show_user_permission(self, command):
-        user_name_tree: Tree = command['user_name']
+        user_name_tree: Tree = command["user_name"]
         user_name_str: str = user_name_tree.children[0].strip("'\"")
         print(f"show_user_permission user_name: {user_name_str}")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/users/{user_name_str}/permission'
+        url = f"http://{self.host}:{self.port}/api/v1/admin/users/{user_name_str}/permission"
         response = self.session.get(url)
         res_json = response.json()
         if response.status_code == 200:
-            self._print_table_simple(res_json['data'])
+            self._print_table_simple(res_json["data"])
         else:
-            print(
-                f"Fail to show user: {user_name_str} permission, code: {res_json['code']}, message: {res_json['message']}")
+            print(f"Fail to show user: {user_name_str} permission, code: {res_json['code']}, message: {res_json['message']}")
 
     def _show_version(self, command):
         print("show_version")
-        url = f'http://{self.host}:{self.port}/api/v1/admin/version'
+        url = f"http://{self.host}:{self.port}/api/v1/admin/version"
         response = self.session.get(url)
         res_json = response.json()
         if response.status_code == 200:
-            self._print_table_simple(res_json['data'])
+            self._print_table_simple(res_json["data"])
         else:
             print(f"Fail to show version, code: {res_json['code']}, message: {res_json['message']}")
 
     def _handle_meta_command(self, command):
-        meta_command = command['command']
-        args = command.get('args', [])
+        meta_command = command["command"]
+        args = command.get("args", [])
 
-        if meta_command in ['?', 'h', 'help']:
+        if meta_command in ["?", "h", "help"]:
             self.show_help()
-        elif meta_command in ['q', 'quit', 'exit']:
+        elif meta_command in ["q", "quit", "exit"]:
             print("Goodbye!")
         else:
             print(f"Meta command '{meta_command}' with args {args}")
@@ -950,16 +910,16 @@ def main():
     cli = AdminCLI()
 
     args = cli.parse_connection_args(sys.argv)
-    if 'error' in args:
+    if "error" in args:
         print("Error: Invalid connection arguments")
         return
 
-    if 'command' in args:
-        if 'password' not in args:
+    if "command" in args:
+        if "password" not in args:
             print("Error: password is missing")
             return
         if cli.verify_admin(args, single_command=True):
-            command: str = args['command']
+            command: str = args["command"]
             # print(f"Run single command: {command}")
             cli.run_single_command(command)
     else:
@@ -974,5 +934,5 @@ def main():
             cli.cmdloop()
 
 
-if __name__ == '__main__':
+if __name__ == "__main__":
     main()

admin/client/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ragflow-cli"
-version = "0.22.1"
+version = "0.23.0"
 description = "Admin Service's client of [RAGFlow](https://github.com/infiniflow/ragflow). The Admin Service provides user management and system monitoring. "
 authors = [{ name = "Lynn", email = "lynn_inf@hotmail.com" }]
 license = { text = "Apache License, Version 2.0" }

admin/client/uv.lock

@@ -196,7 +196,7 @@ wheels = [
 
 [[package]]
 name = "ragflow-cli"
-version = "0.22.1"
+version = "0.23.0"
 source = { virtual = "." }
 dependencies = [
     { name = "beartype" },

admin/server/routes.py

@@ -29,6 +29,11 @@ from common.versions import get_ragflow_version
 admin_bp = Blueprint('admin', __name__, url_prefix='/api/v1/admin')
 
 
+@admin_bp.route('/ping', methods=['GET'])
+def ping():
+    return success_response('PONG')
+
+
 @admin_bp.route('/login', methods=['POST'])
 def login():
     if not request.json:

agent/canvas.py

@@ -278,7 +278,7 @@ class Graph:
 
 class Canvas(Graph):
 
-    def __init__(self, dsl: str, tenant_id=None, task_id=None):
+    def __init__(self, dsl: str, tenant_id=None, task_id=None, canvas_id=None):
         self.globals = {
             "sys.query": "",
             "sys.user_id": tenant_id,
@@ -287,6 +287,7 @@ class Canvas(Graph):
         }
         self.variables = {}
         super().__init__(dsl, tenant_id, task_id)
+        self._id = canvas_id
 
     def load(self):
         super().load()
@@ -721,6 +722,9 @@ class Canvas(Graph):
     def get_mode(self):
         return self.components["begin"]["obj"]._param.mode
 
+    def get_sys_query(self):
+        return self.globals.get("sys.query", "")
+
     def set_global_param(self, **kwargs):
         self.globals.update(kwargs)
 

agent/component/agent_with_tools.py

@@ -86,8 +86,9 @@ class Agent(LLM, ToolBase):
         self.tools = {}
         for idx, cpn in enumerate(self._param.tools):
             cpn = self._load_tool_obj(cpn)
-            name = cpn.get_meta()["function"]["name"]
-            self.tools[f"{name}_{idx}"] = cpn
+            original_name = cpn.get_meta()["function"]["name"]
+            indexed_name = f"{original_name}_{idx}"
+            self.tools[indexed_name] = cpn
 
         self.chat_mdl = LLMBundle(self._canvas.get_tenant_id(), TenantLLMService.llm_id2llm_type(self._param.llm_id), self._param.llm_id,
                                   max_retries=self._param.max_retries,
@@ -95,7 +96,12 @@ class Agent(LLM, ToolBase):
                                   max_rounds=self._param.max_rounds,
                                   verbose_tool_use=True
                                   )
-        self.tool_meta = [v.get_meta() for _,v in self.tools.items()]
+        self.tool_meta = []
+        for indexed_name, tool_obj in self.tools.items():
+            original_meta = tool_obj.get_meta()
+            indexed_meta = deepcopy(original_meta)
+            indexed_meta["function"]["name"] = indexed_name
+            self.tool_meta.append(indexed_meta)
 
         for mcp in self._param.mcp:
             _, mcp_server = MCPServerService.get_by_id(mcp["mcp_id"])
@@ -109,7 +115,8 @@ class Agent(LLM, ToolBase):
 
     def _load_tool_obj(self, cpn: dict) -> object:
         from agent.component import component_class
-        param = component_class(cpn["component_name"] + "Param")()
+        tool_name = cpn["component_name"]
+        param = component_class(tool_name + "Param")()
         param.update(cpn["params"])
         try:
             param.check()
@@ -277,19 +284,15 @@ class Agent(LLM, ToolBase):
         else:
             user_request = history[-1]["content"]
 
-        def build_task_desc(prompt: str, user_request: str, tool_metas: list[dict], user_defined_prompt: dict | None = None) -> str:
+        def build_task_desc(prompt: str, user_request: str, user_defined_prompt: dict | None = None) -> str:
             """Build a minimal task_desc by concatenating prompt, query, and tool schemas."""
             user_defined_prompt = user_defined_prompt or {}
 
-            tools_json = json.dumps(tool_metas, ensure_ascii=False, indent=2)
-
             task_desc = (
                 "### Agent Prompt\n"
                 f"{prompt}\n\n"
                 "### User Request\n"
                 f"{user_request}\n\n"
-                "### Tools (schemas)\n"
-                f"{tools_json}\n"
             )
 
             if user_defined_prompt:
@@ -368,7 +371,7 @@ class Agent(LLM, ToolBase):
                 hist.append({"role": "user", "content": content})
 
         st = timer()
-        task_desc = build_task_desc(prompt, user_request, tool_metas, user_defined_prompt)
+        task_desc = build_task_desc(prompt, user_request, user_defined_prompt)
         self.callback("analyze_task", {}, task_desc, elapsed_time=timer()-st)
         for _ in range(self._param.max_rounds + 1):
             if self.check_if_canceled("Agent streaming"):

agent/component/llm.py

@@ -56,7 +56,6 @@ class LLMParam(ComponentParamBase):
         self.check_nonnegative_number(int(self.max_tokens), "[Agent] Max tokens")
         self.check_decimal_float(float(self.top_p), "[Agent] Top P")
         self.check_empty(self.llm_id, "[Agent] LLM")
-        self.check_empty(self.sys_prompt, "[Agent] System prompt")
         self.check_empty(self.prompts, "[Agent] User prompt")
 
     def gen_conf(self):

agent/component/loopitem.py

@@ -113,6 +113,10 @@ class LoopItem(ComponentBase, ABC):
                 return len(var) == 0
             elif operator == "not empty":
                 return len(var) > 0
+        elif var is None:
+            if operator == "empty":
+                return True
+            return False
 
         raise Exception(f"Invalid operator: {operator}")
 

agent/component/message.py

@@ -33,6 +33,8 @@ from common.connection_utils import timeout
 from common.misc_utils import get_uuid
 from common import settings
 
+from api.db.joint_services.memory_message_service import queue_save_to_memory_task
+
 
 class MessageParam(ComponentParamBase):
     """
@@ -166,6 +168,7 @@ class Message(ComponentBase):
 
         self.set_output("content", all_content)
         self._convert_content(all_content)
+        await self._save_to_memory(all_content)
 
     def _is_jinjia2(self, content:str) -> bool:
         patt = [
@@ -198,6 +201,7 @@ class Message(ComponentBase):
 
         self.set_output("content", content)
         self._convert_content(content)
+        self._save_to_memory(content)
 
     def thoughts(self) -> str:
         return ""
@@ -421,3 +425,16 @@ class Message(ComponentBase):
 
         except Exception as e:
             logging.error(f"Error converting content to {self._param.output_format}: {e}")
+
+    async def _save_to_memory(self, content):
+        if not hasattr(self._param, "memory_ids") or not self._param.memory_ids:
+            return True, "No memory selected."
+
+        message_dict = {
+            "user_id": self._canvas._tenant_id,
+            "agent_id": self._canvas._id,
+            "session_id": self._canvas.task_id,
+            "user_input": self._canvas.get_sys_query(),
+            "agent_response": content
+        }
+        return await queue_save_to_memory_task(self._param.memory_ids, message_dict)

agent/tools/retrieval.py

@@ -25,10 +25,12 @@ from api.db.services.document_service import DocumentService
 from common.metadata_utils import apply_meta_data_filter
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.llm_service import LLMBundle
+from api.db.services.memory_service import MemoryService
+from api.db.joint_services import memory_message_service
 from common import settings
 from common.connection_utils import timeout
 from rag.app.tag import label_question
-from rag.prompts.generator import cross_languages, kb_prompt
+from rag.prompts.generator import cross_languages, kb_prompt, memory_prompt
 
 
 class RetrievalParam(ToolParamBase):
@@ -57,6 +59,7 @@ class RetrievalParam(ToolParamBase):
         self.top_n = 8
         self.top_k = 1024
         self.kb_ids = []
+        self.memory_ids = []
         self.kb_vars = []
         self.rerank_id = ""
         self.empty_response = ""
@@ -81,15 +84,7 @@ class RetrievalParam(ToolParamBase):
 class Retrieval(ToolBase, ABC):
     component_name = "Retrieval"
 
-    @timeout(int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 12)))
-    async def _invoke_async(self, **kwargs):
-        if self.check_if_canceled("Retrieval processing"):
-            return
-
-        if not kwargs.get("query"):
-            self.set_output("formalized_content", self._param.empty_response)
-            return
-
+    async def _retrieve_kb(self, query_text: str):
         kb_ids: list[str] = []
         for id in self._param.kb_ids:
             if id.find("@") < 0:
@@ -124,12 +119,12 @@ class Retrieval(ToolBase, ABC):
         if self._param.rerank_id:
             rerank_mdl = LLMBundle(kbs[0].tenant_id, LLMType.RERANK, self._param.rerank_id)
 
-        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)
+        vars = self.get_input_elements_from_text(query_text)
+        vars = {k: o["value"] for k, o in vars.items()}
+        query = self.string_format(query_text, vars)
 
-        doc_ids=[]
-        if self._param.meta_data_filter!={}:
+        doc_ids = []
+        if self._param.meta_data_filter != {}:
             metas = DocumentService.get_meta_by_kbs(kb_ids)
 
             def _resolve_manual_filter(flt: dict) -> dict:
@@ -198,18 +193,20 @@ class Retrieval(ToolBase, ABC):
 
             if self._param.toc_enhance:
                 chat_mdl = LLMBundle(self._canvas._tenant_id, LLMType.CHAT)
-                cks = settings.retriever.retrieval_by_toc(query, kbinfos["chunks"], [kb.tenant_id for kb in kbs], chat_mdl, self._param.top_n)
+                cks = settings.retriever.retrieval_by_toc(query, kbinfos["chunks"], [kb.tenant_id for kb in kbs],
+                                                          chat_mdl, self._param.top_n)
                 if self.check_if_canceled("Retrieval processing"):
                     return
                 if cks:
                     kbinfos["chunks"] = cks
-            kbinfos["chunks"] = settings.retriever.retrieval_by_children(kbinfos["chunks"], [kb.tenant_id for kb in kbs])
+            kbinfos["chunks"] = settings.retriever.retrieval_by_children(kbinfos["chunks"],
+                                                                         [kb.tenant_id for kb in kbs])
             if self._param.use_kg:
                 ck = settings.kg_retriever.retrieval(query,
-                                                       [kb.tenant_id for kb in kbs],
-                                                       kb_ids,
-                                                       embd_mdl,
-                                                       LLMBundle(self._canvas.get_tenant_id(), LLMType.CHAT))
+                                                     [kb.tenant_id for kb in kbs],
+                                                     kb_ids,
+                                                     embd_mdl,
+                                                     LLMBundle(self._canvas.get_tenant_id(), LLMType.CHAT))
                 if self.check_if_canceled("Retrieval processing"):
                     return
                 if ck["content_with_weight"]:
@@ -218,7 +215,8 @@ class Retrieval(ToolBase, ABC):
             kbinfos = {"chunks": [], "doc_aggs": []}
 
         if self._param.use_kg and kbs:
-            ck = settings.kg_retriever.retrieval(query, [kb.tenant_id for kb in kbs], filtered_kb_ids, embd_mdl, LLMBundle(kbs[0].tenant_id, LLMType.CHAT))
+            ck = settings.kg_retriever.retrieval(query, [kb.tenant_id for kb in kbs], filtered_kb_ids, embd_mdl,
+                                                 LLMBundle(kbs[0].tenant_id, LLMType.CHAT))
             if self.check_if_canceled("Retrieval processing"):
                 return
             if ck["content_with_weight"]:
@@ -248,6 +246,54 @@ class Retrieval(ToolBase, ABC):
 
         return form_cnt
 
+    async def _retrieve_memory(self, query_text: str):
+        memory_ids: list[str] = [memory_id for memory_id in self._param.memory_ids]
+        memory_list = MemoryService.get_by_ids(memory_ids)
+        if not memory_list:
+            raise Exception("No memory is selected.")
+
+        embd_names = list({memory.embd_id for memory in memory_list})
+        assert len(embd_names) == 1, "Memory use different embedding models."
+
+        vars = self.get_input_elements_from_text(query_text)
+        vars = {k: o["value"] for k, o in vars.items()}
+        query = self.string_format(query_text, vars)
+        # query message
+        message_list = memory_message_service.query_message({"memory_id": memory_ids}, {
+            "query": query,
+            "similarity_threshold": self._param.similarity_threshold,
+            "keywords_similarity_weight": self._param.keywords_similarity_weight,
+            "top_n": self._param.top_n
+        })
+        if not message_list:
+            self.set_output("formalized_content", self._param.empty_response)
+            return ""
+        formated_content = "\n".join(memory_prompt(message_list, 200000))
+        # set formalized_content output
+        self.set_output("formalized_content", formated_content)
+
+        return formated_content
+
+    @timeout(int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 12)))
+    async def _invoke_async(self, **kwargs):
+        if self.check_if_canceled("Retrieval processing"):
+            return
+        if not kwargs.get("query"):
+            self.set_output("formalized_content", self._param.empty_response)
+            return
+
+        if hasattr(self._param, "retrieval_from") and self._param.retrieval_from == "dataset":
+            return await self._retrieve_kb(kwargs["query"])
+        elif hasattr(self._param, "retrieval_from") and self._param.retrieval_from == "memory":
+            return await self._retrieve_memory(kwargs["query"])
+        elif self._param.kb_ids:
+            return await self._retrieve_kb(kwargs["query"])
+        elif hasattr(self._param, "memory_ids") and self._param.memory_ids:
+            return await self._retrieve_memory(kwargs["query"])
+        else:
+            self.set_output("formalized_content", self._param.empty_response)
+            return
+
     @timeout(int(os.environ.get("COMPONENT_EXEC_TIMEOUT", 12)))
     def _invoke(self, **kwargs):
         return asyncio.run(self._invoke_async(**kwargs))

api/apps/__init__.py

@@ -38,7 +38,6 @@ settings.init_settings()
 
 __all__ = ["app"]
 
-
 app = Quart(__name__)
 app = cors(app, allow_origin="*")
 
@@ -103,6 +102,7 @@ from werkzeug.local import LocalProxy
 T = TypeVar("T")
 P = ParamSpec("P")
 
+
 def _load_user():
     jwt = Serializer(secret_key=settings.SECRET_KEY)
     authorization = request.headers.get("Authorization")
@@ -164,7 +164,7 @@ def login_required(func: Callable[P, Awaitable[T]]) -> Callable[P, Awaitable[T]]
 
     @wraps(func)
     async def wrapper(*args: P.args, **kwargs: P.kwargs) -> T:
-        if not current_user:# or not session.get("_user_id"):
+        if not current_user:  # or not session.get("_user_id"):
             raise Unauthorized()
         else:
             return await current_app.ensure_async(func)(*args, **kwargs)
@@ -228,6 +228,7 @@ def logout_user():
 
     return True
 
+
 def search_pages_path(page_path):
     app_path_list = [
         path for path in page_path.glob("*_app.py") if not path.name.startswith(".")
@@ -274,6 +275,16 @@ client_urls_prefix = [
 ]
 
 
+@app.errorhandler(404)
+async def not_found(error):
+    error_msg: str = f"The requested URL {request.path} was not found"
+    logging.error(error_msg)
+    return {
+        "error": "Not Found",
+        "message": error_msg,
+    }, 404
+
+
 @app.teardown_request
 def _db_close(exception):
     if exception:

api/apps/canvas_app.py

@@ -153,7 +153,7 @@ async def run():
         return get_json_result(data={"message_id": task_id})
 
     try:
-        canvas = Canvas(cvs.dsl, current_user.id)
+        canvas = Canvas(cvs.dsl, current_user.id, canvas_id=cvs.id)
     except Exception as e:
         return server_error_response(e)
 
@@ -192,7 +192,7 @@ async def rerun():
     if 0 < doc["progress"] < 1:
         return get_data_error_result(message=f"`{doc['name']}` is processing...")
 
-    if settings.docStoreConn.indexExist(search.index_name(current_user.id), doc["kb_id"]):
+    if settings.docStoreConn.index_exist(search.index_name(current_user.id), doc["kb_id"]):
         settings.docStoreConn.delete({"doc_id": doc["id"]}, search.index_name(current_user.id), doc["kb_id"])
     doc["progress_msg"] = ""
     doc["chunk_num"] = 0
@@ -232,7 +232,7 @@ async def reset():
         if not e:
             return get_data_error_result(message="canvas not found.")
 
-        canvas = Canvas(json.dumps(user_canvas.dsl), current_user.id)
+        canvas = Canvas(json.dumps(user_canvas.dsl), current_user.id, canvas_id=user_canvas.id)
         canvas.reset()
         req["dsl"] = json.loads(str(canvas))
         UserCanvasService.update_by_id(req["id"], {"dsl": req["dsl"]})
@@ -270,7 +270,7 @@ def input_form():
                 data=False, message='Only owner of canvas authorized for this operation.',
                 code=RetCode.OPERATING_ERROR)
 
-        canvas = Canvas(json.dumps(user_canvas.dsl), current_user.id)
+        canvas = Canvas(json.dumps(user_canvas.dsl), current_user.id, canvas_id=user_canvas.id)
         return get_json_result(data=canvas.get_component_input_form(cpn_id))
     except Exception as e:
         return server_error_response(e)
@@ -287,7 +287,7 @@ async def debug():
             code=RetCode.OPERATING_ERROR)
     try:
         e, user_canvas = UserCanvasService.get_by_id(req["id"])
-        canvas = Canvas(json.dumps(user_canvas.dsl), current_user.id)
+        canvas = Canvas(json.dumps(user_canvas.dsl), current_user.id, canvas_id=user_canvas.id)
         canvas.reset()
         canvas.message_id = get_uuid()
         component = canvas.get_component(req["component_id"])["obj"]

api/apps/document_app.py

@@ -234,6 +234,10 @@ async def list_docs():
 
     req = await get_request_json()
 
+    return_empty_metadata = req.get("return_empty_metadata", False)
+    if isinstance(return_empty_metadata, str):
+        return_empty_metadata = return_empty_metadata.lower() == "true"
+
     run_status = req.get("run_status", [])
     if run_status:
         invalid_status = {s for s in run_status if s not in VALID_TASK_STATUS}
@@ -248,11 +252,18 @@ async def list_docs():
 
     suffix = req.get("suffix", [])
     metadata_condition = req.get("metadata_condition", {}) or {}
-    if metadata_condition and not isinstance(metadata_condition, dict):
-        return get_data_error_result(message="metadata_condition must be an object.")
     metadata = req.get("metadata", {}) or {}
-    if metadata and not isinstance(metadata, dict):
-        return get_data_error_result(message="metadata must be an object.")
+    if isinstance(metadata, dict) and metadata.get("empty_metadata"):
+        return_empty_metadata = True
+        metadata = {k: v for k, v in metadata.items() if k != "empty_metadata"}
+    if return_empty_metadata:
+        metadata_condition = {}
+        metadata = {}
+    else:
+        if metadata_condition and not isinstance(metadata_condition, dict):
+            return get_data_error_result(message="metadata_condition must be an object.")
+        if metadata and not isinstance(metadata, dict):
+            return get_data_error_result(message="metadata must be an object.")
 
     doc_ids_filter = None
     metas = None
@@ -295,7 +306,19 @@ async def list_docs():
         doc_ids_filter = list(doc_ids_filter)
 
     try:
-        docs, tol = DocumentService.get_by_kb_id(kb_id, page_number, items_per_page, orderby, desc, keywords, run_status, types, suffix, doc_ids_filter)
+        docs, tol = DocumentService.get_by_kb_id(
+            kb_id,
+            page_number,
+            items_per_page,
+            orderby,
+            desc,
+            keywords,
+            run_status,
+            types,
+            suffix,
+            doc_ids_filter,
+            return_empty_metadata=return_empty_metadata,
+        )
 
         if create_time_from or create_time_to:
             filtered_docs = []
@@ -447,6 +470,26 @@ async def metadata_update():
     return get_json_result(data={"updated": updated, "matched_docs": len(target_doc_ids)})
 
 
+@manager.route("/update_metadata_setting", methods=["POST"])  # noqa: F821
+@login_required
+@validate_request("doc_id", "metadata")
+async def update_metadata_setting():
+    req = await get_request_json()
+    if not DocumentService.accessible(req["doc_id"], current_user.id):
+        return get_json_result(data=False, message="No authorization.", code=RetCode.AUTHENTICATION_ERROR)
+
+    e, doc = DocumentService.get_by_id(req["doc_id"])
+    if not e:
+        return get_data_error_result(message="Document not found!")
+
+    DocumentService.update_parser_config(doc.id, {"metadata": req["metadata"]})
+    e, doc = DocumentService.get_by_id(doc.id)
+    if not e:
+        return get_data_error_result(message="Document not found!")
+
+    return get_json_result(data=doc.to_dict())
+
+
 @manager.route("/thumbnails", methods=["GET"])  # noqa: F821
 # @login_required
 def thumbnails():
@@ -564,10 +607,17 @@ async def run():
                 DocumentService.update_by_id(id, info)
                 if req.get("delete", False):
                     TaskService.filter_delete([Task.doc_id == id])
-                    if settings.docStoreConn.indexExist(search.index_name(tenant_id), doc.kb_id):
+                    if settings.docStoreConn.index_exist(search.index_name(tenant_id), doc.kb_id):
                         settings.docStoreConn.delete({"doc_id": id}, search.index_name(tenant_id), doc.kb_id)
 
                 if str(req["run"]) == TaskStatus.RUNNING.value:
+                    if req.get("apply_kb"):
+                        e, kb = KnowledgebaseService.get_by_id(doc.kb_id)
+                        if not e:
+                            raise LookupError("Can't find this dataset!")
+                        doc.parser_config["enable_metadata"] = kb.parser_config.get("enable_metadata", False)
+                        doc.parser_config["metadata"] = kb.parser_config.get("metadata", {})
+                        DocumentService.update_parser_config(doc.id, doc.parser_config)
                     doc_dict = doc.to_dict()
                     DocumentService.run(tenant_id, doc_dict, kb_table_num_map)
 
@@ -615,7 +665,7 @@ async def rename():
                 "title_tks": title_tks,
                 "title_sm_tks": rag_tokenizer.fine_grained_tokenize(title_tks),
             }
-            if settings.docStoreConn.indexExist(search.index_name(tenant_id), doc.kb_id):
+            if settings.docStoreConn.index_exist(search.index_name(tenant_id), doc.kb_id):
                 settings.docStoreConn.update(
                     {"doc_id": req["doc_id"]},
                     es_body,
@@ -696,7 +746,8 @@ async def change_parser():
             tenant_id = DocumentService.get_tenant_id(req["doc_id"])
             if not tenant_id:
                 return get_data_error_result(message="Tenant not found!")
-            if settings.docStoreConn.indexExist(search.index_name(tenant_id), doc.kb_id):
+            DocumentService.delete_chunk_images(doc, tenant_id)
+            if settings.docStoreConn.index_exist(search.index_name(tenant_id), doc.kb_id):
                 settings.docStoreConn.delete({"doc_id": doc.id}, search.index_name(tenant_id), doc.kb_id)
         return None
 

api/apps/kb_app.py

@@ -39,9 +39,9 @@ from api.utils.api_utils import get_json_result
 from rag.nlp import search
 from api.constants import DATASET_NAME_LIMIT
 from rag.utils.redis_conn import REDIS_CONN
-from rag.utils.doc_store_conn import OrderByExpr
 from common.constants import RetCode, PipelineTaskType, StatusEnum, VALID_TASK_STATUS, FileSource, LLMType, PAGERANK_FLD
 from common import settings
+from common.doc_store.doc_store_base import OrderByExpr
 from api.apps import login_required, current_user
 
 
@@ -285,7 +285,7 @@ async def rm():
                     message="Database error (Knowledgebase removal)!")
             for kb in kbs:
                 settings.docStoreConn.delete({"kb_id": kb.id}, search.index_name(kb.tenant_id), kb.id)
-                settings.docStoreConn.deleteIdx(search.index_name(kb.tenant_id), kb.id)
+                settings.docStoreConn.delete_idx(search.index_name(kb.tenant_id), kb.id)
                 if hasattr(settings.STORAGE_IMPL, 'remove_bucket'):
                     settings.STORAGE_IMPL.remove_bucket(kb.id)
             return get_json_result(data=True)
@@ -386,7 +386,7 @@ def knowledge_graph(kb_id):
     }
 
     obj = {"graph": {}, "mind_map": {}}
-    if not settings.docStoreConn.indexExist(search.index_name(kb.tenant_id), kb_id):
+    if not settings.docStoreConn.index_exist(search.index_name(kb.tenant_id), kb_id):
         return get_json_result(data=obj)
     sres = settings.retriever.search(req, search.index_name(kb.tenant_id), [kb_id])
     if not len(sres.ids):
@@ -858,11 +858,11 @@ async def check_embedding():
         index_nm = search.index_name(tenant_id)
 
         res0 = docStoreConn.search(
-            selectFields=[], highlightFields=[],
+            select_fields=[], highlight_fields=[],
             condition={"kb_id": kb_id, "available_int": 1},
-            matchExprs=[], orderBy=OrderByExpr(),
+            match_expressions=[], order_by=OrderByExpr(),
             offset=0, limit=1,
-            indexNames=index_nm, knowledgebaseIds=[kb_id]
+            index_names=index_nm, knowledgebase_ids=[kb_id]
         )
         total = docStoreConn.get_total(res0)
         if total <= 0:
@@ -874,14 +874,14 @@ async def check_embedding():
 
         for off in offsets:
             res1 = docStoreConn.search(
-                selectFields=list(base_fields),
-                highlightFields=[],
+                select_fields=list(base_fields),
+                highlight_fields=[],
                 condition={"kb_id": kb_id, "available_int": 1},
-                matchExprs=[], orderBy=OrderByExpr(),
+                match_expressions=[], order_by=OrderByExpr(),
                 offset=off, limit=1,
-                indexNames=index_nm, knowledgebaseIds=[kb_id]
+                index_names=index_nm, knowledgebase_ids=[kb_id]
             )
-            ids = docStoreConn.get_chunk_ids(res1)
+            ids = docStoreConn.get_doc_ids(res1)
             if not ids:
                 continue
 

api/apps/memories_app.py

@@ -20,10 +20,13 @@ from api.apps import login_required, current_user
 from api.db import TenantPermission
 from api.db.services.memory_service import MemoryService
 from api.db.services.user_service import UserTenantService
-from api.utils.api_utils import validate_request, get_request_json, get_error_argument_result, get_json_result, \
-    not_allowed_parameters
+from api.db.services.canvas_service import UserCanvasService
+from api.db.joint_services.memory_message_service import get_memory_size_cache, judge_system_prompt_is_default
+from api.utils.api_utils import validate_request, get_request_json, get_error_argument_result, get_json_result
 from api.utils.memory_utils import format_ret_data_from_memory, get_memory_type_human
 from api.constants import MEMORY_NAME_LIMIT, MEMORY_SIZE_LIMIT
+from memory.services.messages import MessageService
+from memory.utils.prompt_util import PromptAssembler
 from common.constants import MemoryType, RetCode, ForgettingPolicy
 
 
@@ -57,7 +60,6 @@ async def create_memory():
 
         if res:
             return get_json_result(message=True, data=format_ret_data_from_memory(memory))
-
         else:
             return get_json_result(message=memory, code=RetCode.SERVER_ERROR)
 
@@ -67,7 +69,6 @@ async def create_memory():
 
 @manager.route("/<memory_id>", methods=["PUT"])  # noqa: F821
 @login_required
-@not_allowed_parameters("id", "tenant_id", "memory_type", "storage_type", "embd_id")
 async def update_memory(memory_id):
     req = await get_request_json()
     update_dict = {}
@@ -87,6 +88,14 @@ async def update_memory(memory_id):
         update_dict["permissions"] = req["permissions"]
     if req.get("llm_id"):
         update_dict["llm_id"] = req["llm_id"]
+    if req.get("embd_id"):
+        update_dict["embd_id"] = req["embd_id"]
+    if req.get("memory_type"):
+        memory_type = set(req["memory_type"])
+        invalid_type = memory_type - {e.name.lower() for e in MemoryType}
+        if invalid_type:
+            return get_error_argument_result(f"Memory type '{invalid_type}' is not supported.")
+        update_dict["memory_type"] = list(memory_type)
     # check memory_size valid
     if req.get("memory_size"):
         if not 0 < int(req["memory_size"]) <= MEMORY_SIZE_LIMIT:
@@ -122,9 +131,18 @@ async def update_memory(memory_id):
 
     if not to_update:
         return get_json_result(message=True, data=memory_dict)
+    # check memory empty when update embd_id, memory_type
+    memory_size = get_memory_size_cache(memory_id, current_memory.tenant_id)
+    not_allowed_update = [f for f in ["embd_id", "memory_type"] if f in to_update and memory_size > 0]
+    if not_allowed_update:
+        return get_error_argument_result(f"Can't update {not_allowed_update} when memory isn't empty.")
+    if "memory_type" in to_update:
+        if "system_prompt" not in to_update and judge_system_prompt_is_default(current_memory.system_prompt, current_memory.memory_type):
+            # update old default prompt, assemble a new one
+            to_update["system_prompt"] = PromptAssembler.assemble_system_prompt({"memory_type": to_update["memory_type"]})
 
     try:
-        MemoryService.update_memory(memory_id, to_update)
+        MemoryService.update_memory(current_memory.tenant_id, memory_id, to_update)
         updated_memory = MemoryService.get_by_memory_id(memory_id)
         return get_json_result(message=True, data=format_ret_data_from_memory(updated_memory))
 
@@ -133,7 +151,7 @@ async def update_memory(memory_id):
         return get_json_result(message=str(e), code=RetCode.SERVER_ERROR)
 
 
-@manager.route("/<memory_id>", methods=["DELETE"]) # noqa: F821
+@manager.route("/<memory_id>", methods=["DELETE"])  # noqa: F821
 @login_required
 async def delete_memory(memory_id):
     memory = MemoryService.get_by_memory_id(memory_id)
@@ -141,13 +159,15 @@ async def delete_memory(memory_id):
         return get_json_result(message=True, code=RetCode.NOT_FOUND)
     try:
         MemoryService.delete_memory(memory_id)
+        if MessageService.has_index(memory.tenant_id, memory_id):
+            MessageService.delete_message({"memory_id": memory_id}, memory.tenant_id, memory_id)
         return get_json_result(message=True)
     except Exception as e:
         logging.error(e)
         return get_json_result(message=str(e), code=RetCode.SERVER_ERROR)
 
 
-@manager.route("", methods=["GET"]) # noqa: F821
+@manager.route("", methods=["GET"])  # noqa: F821
 @login_required
 async def list_memory():
     args = request.args
@@ -183,3 +203,26 @@ async def get_memory_config(memory_id):
     if not memory:
         return get_json_result(code=RetCode.NOT_FOUND, message=f"Memory '{memory_id}' not found.")
     return get_json_result(message=True, data=format_ret_data_from_memory(memory))
+
+
+@manager.route("/<memory_id>", methods=["GET"])  # noqa: F821
+@login_required
+async def get_memory_detail(memory_id):
+    args = request.args
+    agent_ids = args.getlist("agent_id")
+    keywords = args.get("keywords", "")
+    keywords = keywords.strip()
+    page = int(args.get("page", 1))
+    page_size = int(args.get("page_size", 50))
+    memory = MemoryService.get_by_memory_id(memory_id)
+    if not memory:
+        return get_json_result(code=RetCode.NOT_FOUND, message=f"Memory '{memory_id}' not found.")
+    messages = MessageService.list_message(
+        memory.tenant_id, memory_id, agent_ids, keywords, page, page_size)
+    agent_name_mapping = {}
+    if messages["message_list"]:
+        agent_list = UserCanvasService.get_basic_info_by_canvas_ids([message["agent_id"] for message in messages["message_list"]])
+        agent_name_mapping = {agent["id"]: agent["title"] for agent in agent_list}
+    for message in messages["message_list"]:
+        message["agent_name"] = agent_name_mapping.get(message["agent_id"], "Unknown")
+    return get_json_result(data={"messages": messages, "storage_type": memory.storage_type}, message=True)

api/apps/messages_app.py

@@ -0,0 +1,168 @@
+#
+#  Copyright 2025 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.
+#
+from quart import request
+from api.apps import login_required
+from api.db.services.memory_service import MemoryService
+from common.time_utils import current_timestamp, timestamp_to_date
+
+from memory.services.messages import MessageService
+from api.db.joint_services import memory_message_service
+from api.utils.api_utils import validate_request, get_request_json, get_error_argument_result, get_json_result
+from common.constants import RetCode
+
+
+@manager.route("", methods=["POST"]) # noqa: F821
+@login_required
+@validate_request("memory_id", "agent_id", "session_id", "user_input", "agent_response")
+async def add_message():
+
+    req = await get_request_json()
+    memory_ids = req["memory_id"]
+    agent_id = req["agent_id"]
+    session_id = req["session_id"]
+    user_id = req["user_id"] if req.get("user_id") else ""
+    user_input = req["user_input"]
+    agent_response = req["agent_response"]
+
+    res = []
+    for memory_id in memory_ids:
+        success, msg = await memory_message_service.save_to_memory(
+            memory_id,
+            {
+                "user_id": user_id,
+                "agent_id": agent_id,
+                "session_id": session_id,
+                "user_input": user_input,
+                "agent_response": agent_response
+            }
+        )
+        res.append({
+            "memory_id": memory_id,
+            "success": success,
+            "message": msg
+        })
+
+    if all([r["success"] for r in res]):
+        return get_json_result(message="Successfully added to memories.")
+
+    return get_json_result(code=RetCode.SERVER_ERROR, message="Some messages failed to add.", data=res)
+
+
+@manager.route("/<memory_id>:<message_id>", methods=["DELETE"]) # noqa: F821
+@login_required
+async def forget_message(memory_id: str, message_id: int):
+
+    memory = MemoryService.get_by_memory_id(memory_id)
+    if not memory:
+        return get_json_result(code=RetCode.NOT_FOUND, message=f"Memory '{memory_id}' not found.")
+
+    forget_time = timestamp_to_date(current_timestamp())
+    update_succeed = MessageService.update_message(
+        {"memory_id": memory_id, "message_id": int(message_id)},
+        {"forget_at": forget_time},
+        memory.tenant_id, memory_id)
+    if update_succeed:
+        return get_json_result(message=update_succeed)
+    else:
+        return get_json_result(code=RetCode.SERVER_ERROR, message=f"Failed to forget message '{message_id}' in memory '{memory_id}'.")
+
+
+@manager.route("/<memory_id>:<message_id>", methods=["PUT"]) # noqa: F821
+@login_required
+@validate_request("status")
+async def update_message(memory_id: str, message_id: int):
+    req = await get_request_json()
+    status = req["status"]
+    if not isinstance(status, bool):
+        return get_error_argument_result("Status must be a boolean.")
+
+    memory = MemoryService.get_by_memory_id(memory_id)
+    if not memory:
+        return get_json_result(code=RetCode.NOT_FOUND, message=f"Memory '{memory_id}' not found.")
+
+    update_succeed = MessageService.update_message({"memory_id": memory_id, "message_id": int(message_id)}, {"status": status}, memory.tenant_id, memory_id)
+    if update_succeed:
+        return get_json_result(message=update_succeed)
+    else:
+        return get_json_result(code=RetCode.SERVER_ERROR, message=f"Failed to set status for message '{message_id}' in memory '{memory_id}'.")
+
+
+@manager.route("/search", methods=["GET"]) # noqa: F821
+@login_required
+async def search_message():
+    args = request.args
+    print(args, flush=True)
+    empty_fields = [f for f in ["memory_id", "query"] if not args.get(f)]
+    if empty_fields:
+        return get_error_argument_result(f"{', '.join(empty_fields)} can't be empty.")
+
+    memory_ids = args.getlist("memory_id")
+    query = args.get("query")
+    similarity_threshold = float(args.get("similarity_threshold", 0.2))
+    keywords_similarity_weight = float(args.get("keywords_similarity_weight", 0.7))
+    top_n = int(args.get("top_n", 5))
+    agent_id = args.get("agent_id", "")
+    session_id = args.get("session_id", "")
+
+    filter_dict = {
+        "memory_id": memory_ids,
+        "agent_id": agent_id,
+        "session_id": session_id
+    }
+    params = {
+        "query": query,
+        "similarity_threshold": similarity_threshold,
+        "keywords_similarity_weight": keywords_similarity_weight,
+        "top_n": top_n
+    }
+    res = memory_message_service.query_message(filter_dict, params)
+    return get_json_result(message=True, data=res)
+
+
+@manager.route("", methods=["GET"]) # noqa: F821
+@login_required
+async def get_messages():
+    args = request.args
+    memory_ids = args.getlist("memory_id")
+    agent_id = args.get("agent_id", "")
+    session_id = args.get("session_id", "")
+    limit = int(args.get("limit", 10))
+    if not memory_ids:
+        return get_error_argument_result("memory_ids is required.")
+    memory_list = MemoryService.get_by_ids(memory_ids)
+    uids = [memory.tenant_id for memory in memory_list]
+    res = MessageService.get_recent_messages(
+        uids,
+        memory_ids,
+        agent_id,
+        session_id,
+        limit
+    )
+    return get_json_result(message=True, data=res)
+
+
+@manager.route("/<memory_id>:<message_id>/content", methods=["GET"]) # noqa: F821
+@login_required
+async def get_message_content(memory_id:str, message_id: int):
+    memory = MemoryService.get_by_memory_id(memory_id)
+    if not memory:
+        return get_json_result(code=RetCode.NOT_FOUND, message=f"Memory '{memory_id}' not found.")
+
+    res = MessageService.get_by_message_id(memory_id, message_id, memory.tenant_id)
+    if res:
+        return get_json_result(message=True, data=res)
+    else:
+        return get_json_result(code=RetCode.NOT_FOUND, message=f"Message '{message_id}' in memory '{memory_id}' not found.")

api/apps/sdk/agents.py

@@ -394,7 +394,7 @@ async def webhook(agent_id: str):
     if not isinstance(cvs.dsl, str):
         dsl = json.dumps(cvs.dsl, ensure_ascii=False)
     try:
-        canvas = Canvas(dsl, cvs.user_id, agent_id)
+        canvas = Canvas(dsl, cvs.user_id, agent_id, canvas_id=agent_id)
     except Exception as e:
         resp=get_data_error_result(code=RetCode.BAD_REQUEST,message=str(e))
         resp.status_code = RetCode.BAD_REQUEST

api/apps/sdk/dataset.py

@@ -495,7 +495,7 @@ def knowledge_graph(tenant_id, dataset_id):
     }
 
     obj = {"graph": {}, "mind_map": {}}
-    if not settings.docStoreConn.indexExist(search.index_name(kb.tenant_id), dataset_id):
+    if not settings.docStoreConn.index_exist(search.index_name(kb.tenant_id), dataset_id):
         return get_result(data=obj)
     sres = settings.retriever.search(req, search.index_name(kb.tenant_id), [dataset_id])
     if not len(sres.ids):

api/apps/sdk/doc.py

@@ -1080,7 +1080,7 @@ def list_chunks(tenant_id, dataset_id, document_id):
         res["chunks"].append(final_chunk)
         _ = Chunk(**final_chunk)
 
-    elif settings.docStoreConn.indexExist(search.index_name(tenant_id), dataset_id):
+    elif settings.docStoreConn.index_exist(search.index_name(tenant_id), dataset_id):
         sres = settings.retriever.search(query, search.index_name(tenant_id), [dataset_id], emb_mdl=None, highlight=True)
         res["total"] = sres.total
         for id in sres.ids:
@@ -1286,6 +1286,9 @@ async def rm_chunk(tenant_id, dataset_id, document_id):
     if "chunk_ids" in req:
         unique_chunk_ids, duplicate_messages = check_duplicate_ids(req["chunk_ids"], "chunk")
         condition["id"] = unique_chunk_ids
+    else:
+        unique_chunk_ids = []
+        duplicate_messages = []
     chunk_number = settings.docStoreConn.delete(condition, search.index_name(tenant_id), dataset_id)
     if chunk_number != 0:
         DocumentService.decrement_chunk_num(document_id, dataset_id, 1, chunk_number, 0)

api/apps/sdk/session.py

@@ -88,7 +88,7 @@ async def create_agent_session(tenant_id, agent_id):
         cvs.dsl = json.dumps(cvs.dsl, ensure_ascii=False)
 
     session_id = get_uuid()
-    canvas = Canvas(cvs.dsl, tenant_id, agent_id)
+    canvas = Canvas(cvs.dsl, tenant_id, agent_id, canvas_id=cvs.id)
     canvas.reset()
 
     cvs.dsl = json.loads(str(canvas))
@@ -986,7 +986,7 @@ async def begin_inputs(agent_id):
     if not e:
         return get_error_data_result(f"Can't find agent by ID: {agent_id}")
 
-    canvas = Canvas(json.dumps(cvs.dsl), objs[0].tenant_id)
+    canvas = Canvas(json.dumps(cvs.dsl), objs[0].tenant_id, canvas_id=cvs.id)
     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()})

api/apps/system_app.py

@@ -177,7 +177,7 @@ def healthz():
     return jsonify(result), (200 if all_ok else 500)
 
 
-@manager.route("/ping", methods=["GET"]) # noqa: F821
+@manager.route("/ping", methods=["GET"])  # noqa: F821
 def ping():
     return "pong", 200
 
@@ -213,7 +213,7 @@ def new_token():
         if not tenants:
             return get_data_error_result(message="Tenant not found!")
 
-        tenant_id = [tenant for tenant in tenants if tenant.role == 'owner'][0].tenant_id
+        tenant_id = [tenant for tenant in tenants if tenant.role == "owner"][0].tenant_id
         obj = {
             "tenant_id": tenant_id,
             "token": generate_confirmation_token(),
@@ -268,13 +268,12 @@ def token_list():
         if not tenants:
             return get_data_error_result(message="Tenant not found!")
 
-        tenant_id = [tenant for tenant in tenants if tenant.role == 'owner'][0].tenant_id
+        tenant_id = [tenant for tenant in tenants if tenant.role == "owner"][0].tenant_id
         objs = APITokenService.query(tenant_id=tenant_id)
         objs = [o.to_dict() for o in objs]
         for o in objs:
             if not o["beta"]:
-                o["beta"] = generate_confirmation_token().replace(
-                    "ragflow-", "")[:32]
+                o["beta"] = generate_confirmation_token().replace("ragflow-", "")[:32]
                 APITokenService.filter_update([APIToken.tenant_id == tenant_id, APIToken.token == o["token"]], o)
         return get_json_result(data=objs)
     except Exception as e:
@@ -307,13 +306,19 @@ def rm(token):
               type: boolean
               description: Deletion status.
     """
-    APITokenService.filter_delete(
-        [APIToken.tenant_id == current_user.id, APIToken.token == token]
-    )
-    return get_json_result(data=True)
+    try:
+        tenants = UserTenantService.query(user_id=current_user.id)
+        if not tenants:
+            return get_data_error_result(message="Tenant not found!")
+
+        tenant_id = tenants[0].tenant_id
+        APITokenService.filter_delete([APIToken.tenant_id == tenant_id, APIToken.token == token])
+        return get_json_result(data=True)
+    except Exception as e:
+        return server_error_response(e)
 
 
-@manager.route('/config', methods=['GET'])  # noqa: F821
+@manager.route("/config", methods=["GET"])  # noqa: F821
 def get_config():
     """
     Get system configuration.
@@ -330,6 +335,4 @@ def get_config():
                         type: integer 0 means disabled, 1 means enabled
                         description: Whether user registration is enabled
     """
-    return get_json_result(data={
-        "registerEnabled": settings.REGISTER_ENABLED
-    })
+    return get_json_result(data={"registerEnabled": settings.REGISTER_ENABLED})

api/db/db_models.py

@@ -1189,7 +1189,7 @@ class Memory(DataBaseModel):
     permissions = CharField(max_length=16, null=False, index=True, help_text="me|team", default="me")
     description = TextField(null=True, help_text="description")
     memory_size = IntegerField(default=5242880, null=False, index=False)
-    forgetting_policy = CharField(max_length=32, null=False, default="fifo", index=False, help_text="lru|fifo")
+    forgetting_policy = CharField(max_length=32, null=False, default="FIFO", index=False, help_text="LRU|FIFO")
     temperature = FloatField(default=0.5, index=False)
     system_prompt = TextField(null=True, help_text="system prompt", index=False)
     user_prompt = TextField(null=True, help_text="user prompt", index=False)

api/db/init_data.py

@@ -30,6 +30,7 @@ from api.db.services.knowledgebase_service import KnowledgebaseService
 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.db.joint_services.memory_message_service import init_message_id_sequence, init_memory_size_cache
 from common.constants import LLMType
 from common.file_utils import get_project_base_directory
 from common import settings
@@ -169,6 +170,8 @@ def init_web_data():
     #    init_superuser()
 
     add_graph_templates()
+    init_message_id_sequence()
+    init_memory_size_cache()
     logging.info("init web data success:{}".format(time.time() - start_time))
 
 

api/db/joint_services/memory_message_service.py

@@ -0,0 +1,389 @@
+#
+#  Copyright 2025 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 typing import List
+
+from api.db.services.task_service import TaskService
+from common import settings
+from common.time_utils import current_timestamp, timestamp_to_date, format_iso_8601_to_ymd_hms
+from common.constants import MemoryType, LLMType
+from common.doc_store.doc_store_base import FusionExpr
+from common.misc_utils import get_uuid
+from api.db.db_utils import bulk_insert_into_db
+from api.db.db_models import Task
+from api.db.services.memory_service import MemoryService
+from api.db.services.tenant_llm_service import TenantLLMService
+from api.db.services.llm_service import LLMBundle
+from api.utils.memory_utils import get_memory_type_human
+from memory.services.messages import MessageService
+from memory.services.query import MsgTextQuery, get_vector
+from memory.utils.prompt_util import PromptAssembler
+from memory.utils.msg_util import get_json_result_from_llm_response
+from rag.utils.redis_conn import REDIS_CONN
+
+
+async def save_to_memory(memory_id: str, message_dict: dict):
+    """
+    :param memory_id:
+    :param message_dict: {
+        "user_id": str,
+        "agent_id": str,
+        "session_id": str,
+        "user_input": str,
+        "agent_response": str
+    }
+    """
+    memory = MemoryService.get_by_memory_id(memory_id)
+    if not memory:
+        return False, f"Memory '{memory_id}' not found."
+
+    tenant_id = memory.tenant_id
+    extracted_content = await extract_by_llm(
+        tenant_id,
+        memory.llm_id,
+        {"temperature": memory.temperature},
+        get_memory_type_human(memory.memory_type),
+        message_dict.get("user_input", ""),
+        message_dict.get("agent_response", "")
+    ) if memory.memory_type != MemoryType.RAW.value else []  # if only RAW, no need to extract
+    raw_message_id = REDIS_CONN.generate_auto_increment_id(namespace="memory")
+    message_list = [{
+        "message_id": raw_message_id,
+        "message_type": MemoryType.RAW.name.lower(),
+        "source_id": 0,
+        "memory_id": memory_id,
+        "user_id": "",
+        "agent_id": message_dict["agent_id"],
+        "session_id": message_dict["session_id"],
+        "content": f"User Input: {message_dict.get('user_input')}\nAgent Response: {message_dict.get('agent_response')}",
+        "valid_at": timestamp_to_date(current_timestamp()),
+        "invalid_at": None,
+        "forget_at": None,
+        "status": True
+    }, *[{
+        "message_id": REDIS_CONN.generate_auto_increment_id(namespace="memory"),
+        "message_type": content["message_type"],
+        "source_id": raw_message_id,
+        "memory_id": memory_id,
+        "user_id": "",
+        "agent_id": message_dict["agent_id"],
+        "session_id": message_dict["session_id"],
+        "content": content["content"],
+        "valid_at": content["valid_at"],
+        "invalid_at": content["invalid_at"] if content["invalid_at"] else None,
+        "forget_at": None,
+        "status": True
+    } for content in extracted_content]]
+    return await embed_and_save(memory, message_list)
+
+
+async def save_extracted_to_memory_only(memory_id: str, message_dict, source_message_id: int):
+    memory = MemoryService.get_by_memory_id(memory_id)
+    if not memory:
+        return False, f"Memory '{memory_id}' not found."
+
+    if memory.memory_type == MemoryType.RAW.value:
+        return True, f"Memory '{memory_id}' don't need to extract."
+
+    tenant_id = memory.tenant_id
+    extracted_content = await extract_by_llm(
+        tenant_id,
+        memory.llm_id,
+        {"temperature": memory.temperature},
+        get_memory_type_human(memory.memory_type),
+        message_dict.get("user_input", ""),
+        message_dict.get("agent_response", "")
+    )
+    message_list = [{
+        "message_id": REDIS_CONN.generate_auto_increment_id(namespace="memory"),
+        "message_type": content["message_type"],
+        "source_id": source_message_id,
+        "memory_id": memory_id,
+        "user_id": "",
+        "agent_id": message_dict["agent_id"],
+        "session_id": message_dict["session_id"],
+        "content": content["content"],
+        "valid_at": content["valid_at"],
+        "invalid_at": content["invalid_at"] if content["invalid_at"] else None,
+        "forget_at": None,
+        "status": True
+    } for content in extracted_content]
+    if not message_list:
+        return True, "No memory extracted from raw message."
+
+    return await embed_and_save(memory, message_list)
+
+
+async def extract_by_llm(tenant_id: str, llm_id: str, extract_conf: dict, memory_type: List[str], user_input: str,
+                         agent_response: str, system_prompt: str = "", user_prompt: str="") -> List[dict]:
+    llm_type = TenantLLMService.llm_id2llm_type(llm_id)
+    if not llm_type:
+        raise RuntimeError(f"Unknown type of LLM '{llm_id}'")
+    if not system_prompt:
+        system_prompt = PromptAssembler.assemble_system_prompt({"memory_type": memory_type})
+    conversation_content = f"User Input: {user_input}\nAgent Response: {agent_response}"
+    conversation_time = timestamp_to_date(current_timestamp())
+    user_prompts = []
+    if user_prompt:
+        user_prompts.append({"role": "user", "content": user_prompt})
+        user_prompts.append({"role": "user", "content": f"Conversation: {conversation_content}\nConversation Time: {conversation_time}\nCurrent Time: {conversation_time}"})
+    else:
+        user_prompts.append({"role": "user", "content": PromptAssembler.assemble_user_prompt(conversation_content, conversation_time, conversation_time)})
+    llm = LLMBundle(tenant_id, llm_type, llm_id)
+    res = await llm.async_chat(system_prompt, user_prompts, extract_conf)
+    res_json = get_json_result_from_llm_response(res)
+    return [{
+        "content": extracted_content["content"],
+        "valid_at": format_iso_8601_to_ymd_hms(extracted_content["valid_at"]),
+        "invalid_at": format_iso_8601_to_ymd_hms(extracted_content["invalid_at"]) if extracted_content.get("invalid_at") else "",
+        "message_type": message_type
+    } for message_type, extracted_content_list in res_json.items() for extracted_content in extracted_content_list]
+
+
+async def embed_and_save(memory, message_list: list[dict]):
+    embedding_model = LLMBundle(memory.tenant_id, llm_type=LLMType.EMBEDDING, llm_name=memory.embd_id)
+    vector_list, _ = embedding_model.encode([msg["content"] for msg in message_list])
+    for idx, msg in enumerate(message_list):
+        msg["content_embed"] = vector_list[idx]
+    vector_dimension = len(vector_list[0])
+    if not MessageService.has_index(memory.tenant_id, memory.id):
+        created = MessageService.create_index(memory.tenant_id, memory.id, vector_size=vector_dimension)
+        if not created:
+            return False, "Failed to create message index."
+
+    new_msg_size = sum([MessageService.calculate_message_size(m) for m in message_list])
+    current_memory_size = get_memory_size_cache(memory.tenant_id, memory.id)
+    if new_msg_size + current_memory_size > memory.memory_size:
+        size_to_delete = current_memory_size + new_msg_size - memory.memory_size
+        if memory.forgetting_policy == "FIFO":
+            message_ids_to_delete, delete_size = MessageService.pick_messages_to_delete_by_fifo(memory.id, memory.tenant_id,
+                                                                                                size_to_delete)
+            MessageService.delete_message({"message_id": message_ids_to_delete}, memory.tenant_id, memory.id)
+            decrease_memory_size_cache(memory.id, delete_size)
+        else:
+            return False, "Failed to insert message into memory. Memory size reached limit and cannot decide which to delete."
+    fail_cases = MessageService.insert_message(message_list, memory.tenant_id, memory.id)
+    if fail_cases:
+        return False, "Failed to insert message into memory. Details: " + "; ".join(fail_cases)
+
+    increase_memory_size_cache(memory.id, new_msg_size)
+    return True, "Message saved successfully."
+
+
+def query_message(filter_dict: dict, params: dict):
+    """
+    :param filter_dict: {
+        "memory_id": List[str],
+        "agent_id": optional
+        "session_id": optional
+    }
+    :param params: {
+        "query": question str,
+        "similarity_threshold": float,
+        "keywords_similarity_weight": float,
+        "top_n": int
+    }
+    """
+    memory_ids = filter_dict["memory_id"]
+    memory_list = MemoryService.get_by_ids(memory_ids)
+    if not memory_list:
+        return []
+
+    condition_dict = {k: v for k, v in filter_dict.items() if v}
+    uids = [memory.tenant_id for memory in memory_list]
+
+    question = params["query"]
+    question = question.strip()
+    memory = memory_list[0]
+    embd_model = LLMBundle(memory.tenant_id, llm_type=LLMType.EMBEDDING, llm_name=memory.embd_id)
+    match_dense = get_vector(question, embd_model, similarity=params["similarity_threshold"])
+    match_text, _ = MsgTextQuery().question(question, min_match=params["similarity_threshold"])
+    keywords_similarity_weight = params.get("keywords_similarity_weight", 0.7)
+    fusion_expr = FusionExpr("weighted_sum", params["top_n"], {"weights": ",".join([str(1 - keywords_similarity_weight), str(keywords_similarity_weight)])})
+
+    return MessageService.search_message(memory_ids, condition_dict, uids, [match_text, match_dense, fusion_expr], params["top_n"])
+
+
+def init_message_id_sequence():
+    message_id_redis_key = "id_generator:memory"
+    if REDIS_CONN.exist(message_id_redis_key):
+        current_max_id = REDIS_CONN.get(message_id_redis_key)
+        logging.info(f"No need to init message_id sequence, current max id is {current_max_id}.")
+    else:
+        max_id = 1
+        exist_memory_list = MemoryService.get_all_memory()
+        if not exist_memory_list:
+            REDIS_CONN.set(message_id_redis_key, max_id)
+        else:
+            max_id = MessageService.get_max_message_id(
+                uid_list=[m.tenant_id for m in exist_memory_list],
+                memory_ids=[m.id for m in exist_memory_list]
+            )
+            REDIS_CONN.set(message_id_redis_key, max_id)
+        logging.info(f"Init message_id sequence done, current max id is {max_id}.")
+
+
+def get_memory_size_cache(memory_id: str, uid: str):
+    redis_key = f"memory_{memory_id}"
+    if REDIS_CONN.exist(redis_key):
+        return int(REDIS_CONN.get(redis_key))
+    else:
+        memory_size_map = MessageService.calculate_memory_size(
+            [memory_id],
+            [uid]
+        )
+        memory_size = memory_size_map.get(memory_id, 0)
+        set_memory_size_cache(memory_id, memory_size)
+        return memory_size
+
+
+def set_memory_size_cache(memory_id: str, size: int):
+    redis_key = f"memory_{memory_id}"
+    return REDIS_CONN.set(redis_key, size)
+
+
+def increase_memory_size_cache(memory_id: str, size: int):
+    redis_key = f"memory_{memory_id}"
+    return REDIS_CONN.incrby(redis_key, size)
+
+
+def decrease_memory_size_cache(memory_id: str, size: int):
+    redis_key = f"memory_{memory_id}"
+    return REDIS_CONN.decrby(redis_key, size)
+
+
+def init_memory_size_cache():
+    memory_list = MemoryService.get_all_memory()
+    if not memory_list:
+        logging.info("No memory found, no need to init memory size.")
+    else:
+        for m in memory_list:
+            get_memory_size_cache(m.id, m.tenant_id)
+        logging.info("Memory size cache init done.")
+
+
+def judge_system_prompt_is_default(system_prompt: str, memory_type: int|list[str]):
+    memory_type_list = memory_type if isinstance(memory_type, list) else get_memory_type_human(memory_type)
+    return system_prompt == PromptAssembler.assemble_system_prompt({"memory_type": memory_type_list})
+
+
+async def queue_save_to_memory_task(memory_ids: list[str], message_dict: dict):
+    """
+    :param memory_ids:
+    :param message_dict: {
+        "user_id": str,
+        "agent_id": str,
+        "session_id": str,
+        "user_input": str,
+        "agent_response": str
+    }
+    """
+    def new_task(_memory_id: str, _source_id: int):
+        return {
+            "id": get_uuid(),
+            "doc_id": _memory_id,
+            "task_type": "memory",
+            "progress": 0.0,
+            "digest": str(_source_id)
+        }
+
+    not_found_memory = []
+    failed_memory = []
+    for memory_id in memory_ids:
+        memory = MemoryService.get_by_memory_id(memory_id)
+        if not memory:
+            not_found_memory.append(memory_id)
+            continue
+
+        raw_message_id = REDIS_CONN.generate_auto_increment_id(namespace="memory")
+        raw_message = {
+            "message_id": raw_message_id,
+            "message_type": MemoryType.RAW.name.lower(),
+            "source_id": 0,
+            "memory_id": memory_id,
+            "user_id": "",
+            "agent_id": message_dict["agent_id"],
+            "session_id": message_dict["session_id"],
+            "content": f"User Input: {message_dict.get('user_input')}\nAgent Response: {message_dict.get('agent_response')}",
+            "valid_at": timestamp_to_date(current_timestamp()),
+            "invalid_at": None,
+            "forget_at": None,
+            "status": True
+        }
+        res, msg = await embed_and_save(memory, [raw_message])
+        if not res:
+            failed_memory.append({"memory_id": memory_id, "fail_msg": msg})
+            continue
+
+        task = new_task(memory_id, raw_message_id)
+        bulk_insert_into_db(Task, [task], replace_on_conflict=True)
+        task_message = {
+            "id": task["id"],
+            "task_id": task["id"],
+            "task_type": task["task_type"],
+            "memory_id": memory_id,
+            "source_id": raw_message_id,
+            "message_dict": message_dict
+        }
+        if not REDIS_CONN.queue_product(settings.get_svr_queue_name(priority=0), message=task_message):
+            failed_memory.append({"memory_id": memory_id, "fail_msg": "Can't access Redis."})
+
+    error_msg = ""
+    if not_found_memory:
+        error_msg = f"Memory {not_found_memory} not found."
+    if failed_memory:
+        error_msg += "".join([f"Memory {fm['memory_id']} failed. Detail: {fm['fail_msg']}" for fm in failed_memory])
+
+    if error_msg:
+        return False, error_msg
+
+    return True, "All add to task."
+
+
+async def handle_save_to_memory_task(task_param: dict):
+    """
+    :param task_param: {
+        "id": task_id
+        "memory_id": id
+        "source_id": id
+        "message_dict": {
+            "user_id": str,
+            "agent_id": str,
+            "session_id": str,
+            "user_input": str,
+            "agent_response": str
+        }
+    }
+    """
+    _, task = TaskService.get_by_id(task_param["id"])
+    if not task:
+        return False, f"Task {task_param['id']} is not found."
+    if task.progress == -1:
+        return False, f"Task {task_param['id']} is already failed."
+    now_time = current_timestamp()
+    TaskService.update_by_id(task_param["id"], {"begin_at": timestamp_to_date(now_time)})
+
+    memory_id = task_param["memory_id"]
+    source_id = task_param["source_id"]
+    message_dict = task_param["message_dict"]
+    success, msg = await save_extracted_to_memory_only(memory_id, message_dict, source_id)
+    if success:
+        TaskService.update_progress(task.id, {"progress": 1.0,  "progress_msg": msg})
+        return True, msg
+
+    logging.error(msg)
+    TaskService.update_progress(task.id, {"progress": -1, "progress_msg": None})
+    return False, msg

api/db/joint_services/user_account_service.py

@@ -34,6 +34,8 @@ from api.db.services.task_service import TaskService
 from api.db.services.tenant_llm_service import TenantLLMService
 from api.db.services.user_canvas_version import UserCanvasVersionService
 from api.db.services.user_service import TenantService, UserService, UserTenantService
+from api.db.services.memory_service import MemoryService
+from memory.services.messages import MessageService
 from rag.nlp import search
 from common.constants import ActiveEnum
 from common import settings
@@ -200,7 +202,16 @@ def delete_user_data(user_id: str) -> dict:
             done_msg += f"- Deleted {llm_delete_res} tenant-LLM records.\n"
             langfuse_delete_res = TenantLangfuseService.delete_ty_tenant_id(tenant_id)
             done_msg += f"- Deleted {langfuse_delete_res} langfuse records.\n"
-            # step1.3 delete own tenant
+            # step1.3 delete memory and messages
+            user_memory = MemoryService.get_by_tenant_id(tenant_id)
+            if user_memory:
+                for memory in user_memory:
+                    if MessageService.has_index(tenant_id, memory.id):
+                        MessageService.delete_index(tenant_id, memory.id)
+                done_msg += " Deleted memory index."
+                memory_delete_res = MemoryService.delete_by_ids([m.id for m in user_memory])
+                done_msg += f"Deleted {memory_delete_res} memory datasets."
+            # step1.4 delete own tenant
             tenant_delete_res = TenantService.delete_by_id(tenant_id)
             done_msg += f"- Deleted {tenant_delete_res} tenant.\n"
         # step2 delete user-tenant relation

api/db/services/canvas_service.py

@@ -123,6 +123,19 @@ class UserCanvasService(CommonService):
             logging.exception(e)
             return False, None
 
+    @classmethod
+    @DB.connection_context()
+    def get_basic_info_by_canvas_ids(cls, canvas_id):
+        fields = [
+            cls.model.id,
+            cls.model.avatar,
+            cls.model.user_id,
+            cls.model.title,
+            cls.model.permission,
+            cls.model.canvas_category
+        ]
+        return cls.model.select(*fields).where(cls.model.id.in_(canvas_id)).dicts()
+
     @classmethod
     @DB.connection_context()
     def get_by_tenant_ids(cls, joined_tenant_ids, user_id,
@@ -198,7 +211,7 @@ async def completion(tenant_id, agent_id, session_id=None, **kwargs):
         if not isinstance(cvs.dsl, str):
             cvs.dsl = json.dumps(cvs.dsl, ensure_ascii=False)
         session_id=get_uuid()
-        canvas = Canvas(cvs.dsl, tenant_id, agent_id)
+        canvas = Canvas(cvs.dsl, tenant_id, agent_id, canvas_id=cvs.id)
         canvas.reset()
         conv = {
             "id": session_id,

api/db/services/conversation_service.py

@@ -116,6 +116,16 @@ async def async_completion(tenant_id, chat_id, question, name="New session", ses
                                     ensure_ascii=False) + "\n\n"
             yield "data:" + json.dumps({"code": 0, "message": "", "data": True}, ensure_ascii=False) + "\n\n"
             return
+        else:
+            answer = {
+                "answer": conv["message"][0]["content"],
+                "reference": {},
+                "audio_binary": None,
+                "id": None,
+                "session_id": session_id
+            }
+            yield answer
+            return
 
     conv = ConversationService.query(id=session_id, dialog_id=chat_id)
     if not conv:

api/db/services/document_service.py

@@ -33,12 +33,13 @@ from api.db.db_models import DB, Document, Knowledgebase, Task, Tenant, UserTena
 from api.db.db_utils import bulk_insert_into_db
 from api.db.services.common_service import CommonService
 from api.db.services.knowledgebase_service import KnowledgebaseService
+from common.metadata_utils import dedupe_list
 from common.misc_utils import get_uuid
 from common.time_utils import current_timestamp, get_format_time
 from common.constants import LLMType, ParserType, StatusEnum, TaskStatus, SVR_CONSUMER_GROUP_NAME
 from rag.nlp import rag_tokenizer, search
 from rag.utils.redis_conn import REDIS_CONN
-from rag.utils.doc_store_conn import OrderByExpr
+from common.doc_store.doc_store_base import OrderByExpr
 from common import settings
 
 
@@ -124,26 +125,26 @@ class DocumentService(CommonService):
 
     @classmethod
     @DB.connection_context()
-    def get_by_kb_id(cls, kb_id, page_number, items_per_page,
-                     orderby, desc, keywords, run_status, types, suffix, doc_ids=None):
+    def get_by_kb_id(cls, kb_id, page_number, items_per_page, orderby, desc, keywords, run_status, types, suffix, doc_ids=None, return_empty_metadata=False):
         fields = cls.get_cls_model_fields()
         if keywords:
-            docs = cls.model.select(*[*fields, UserCanvas.title.alias("pipeline_name"), User.nickname])\
-                .join(File2Document, on=(File2Document.document_id == cls.model.id))\
-                .join(File, on=(File.id == File2Document.file_id))\
-                .join(UserCanvas, on=(cls.model.pipeline_id == UserCanvas.id), join_type=JOIN.LEFT_OUTER)\
-                .join(User, on=(cls.model.created_by == User.id), join_type=JOIN.LEFT_OUTER)\
-                .where(
-                    (cls.model.kb_id == kb_id),
-                    (fn.LOWER(cls.model.name).contains(keywords.lower()))
-                )
+            docs = (
+                cls.model.select(*[*fields, UserCanvas.title.alias("pipeline_name"), User.nickname])
+                .join(File2Document, on=(File2Document.document_id == cls.model.id))
+                .join(File, on=(File.id == File2Document.file_id))
+                .join(UserCanvas, on=(cls.model.pipeline_id == UserCanvas.id), join_type=JOIN.LEFT_OUTER)
+                .join(User, on=(cls.model.created_by == User.id), join_type=JOIN.LEFT_OUTER)
+                .where((cls.model.kb_id == kb_id), (fn.LOWER(cls.model.name).contains(keywords.lower())))
+            )
         else:
-            docs = cls.model.select(*[*fields, UserCanvas.title.alias("pipeline_name"), User.nickname])\
-                .join(File2Document, on=(File2Document.document_id == cls.model.id))\
-                .join(UserCanvas, on=(cls.model.pipeline_id == UserCanvas.id), join_type=JOIN.LEFT_OUTER)\
-                .join(File, on=(File.id == File2Document.file_id))\
-                .join(User, on=(cls.model.created_by == User.id), join_type=JOIN.LEFT_OUTER)\
+            docs = (
+                cls.model.select(*[*fields, UserCanvas.title.alias("pipeline_name"), User.nickname])
+                .join(File2Document, on=(File2Document.document_id == cls.model.id))
+                .join(UserCanvas, on=(cls.model.pipeline_id == UserCanvas.id), join_type=JOIN.LEFT_OUTER)
+                .join(File, on=(File.id == File2Document.file_id))
+                .join(User, on=(cls.model.created_by == User.id), join_type=JOIN.LEFT_OUTER)
                 .where(cls.model.kb_id == kb_id)
+            )
 
         if doc_ids:
             docs = docs.where(cls.model.id.in_(doc_ids))
@@ -153,6 +154,8 @@ class DocumentService(CommonService):
             docs = docs.where(cls.model.type.in_(types))
         if suffix:
             docs = docs.where(cls.model.suffix.in_(suffix))
+        if return_empty_metadata:
+            docs = docs.where(fn.COALESCE(fn.JSON_LENGTH(cls.model.meta_fields), 0) == 0)
 
         count = docs.count()
         if desc:
@@ -160,7 +163,6 @@ class DocumentService(CommonService):
         else:
             docs = docs.order_by(cls.model.getter_by(orderby).asc())
 
-
         if page_number and items_per_page:
             docs = docs.paginate(page_number, items_per_page)
 
@@ -216,18 +218,16 @@ class DocumentService(CommonService):
         suffix_counter = {}
         run_status_counter = {}
         metadata_counter = {}
+        empty_metadata_count = 0
 
         for row in rows:
             suffix_counter[row.suffix] = suffix_counter.get(row.suffix, 0) + 1
             run_status_counter[str(row.run)] = run_status_counter.get(str(row.run), 0) + 1
             meta_fields = row.meta_fields or {}
-            if isinstance(meta_fields, str):
-                try:
-                    meta_fields = json.loads(meta_fields)
-                except Exception:
-                    meta_fields = {}
-            if not isinstance(meta_fields, dict):
+            if not meta_fields:
+                empty_metadata_count += 1
                 continue
+            has_valid_meta = False
             for key, value in meta_fields.items():
                 values = value if isinstance(value, list) else [value]
                 for vv in values:
@@ -239,7 +239,11 @@ class DocumentService(CommonService):
                     if key not in metadata_counter:
                         metadata_counter[key] = {}
                     metadata_counter[key][sv] = metadata_counter[key].get(sv, 0) + 1
+                    has_valid_meta = True
+            if not has_valid_meta:
+                empty_metadata_count += 1
 
+        metadata_counter["empty_metadata"] = {"true": empty_metadata_count}
         return {
             "suffix": suffix_counter,
             "run_status": run_status_counter,
@@ -338,21 +342,7 @@ class DocumentService(CommonService):
         cls.clear_chunk_num(doc.id)
         try:
             TaskService.filter_delete([Task.doc_id == doc.id])
-            page = 0
-            page_size = 1000
-            all_chunk_ids = []
-            while True:
-                chunks = settings.docStoreConn.search(["img_id"], [], {"doc_id": doc.id}, [], OrderByExpr(),
-                                                      page * page_size, page_size, search.index_name(tenant_id),
-                                                      [doc.kb_id])
-                chunk_ids = settings.docStoreConn.get_chunk_ids(chunks)
-                if not chunk_ids:
-                    break
-                all_chunk_ids.extend(chunk_ids)
-                page += 1
-            for cid in all_chunk_ids:
-                if settings.STORAGE_IMPL.obj_exist(doc.kb_id, cid):
-                    settings.STORAGE_IMPL.rm(doc.kb_id, cid)
+            cls.delete_chunk_images(doc, tenant_id)
             if doc.thumbnail and not doc.thumbnail.startswith(IMG_BASE64_PREFIX):
                 if settings.STORAGE_IMPL.obj_exist(doc.kb_id, doc.thumbnail):
                     settings.STORAGE_IMPL.rm(doc.kb_id, doc.thumbnail)
@@ -374,6 +364,23 @@ class DocumentService(CommonService):
             pass
         return cls.delete_by_id(doc.id)
 
+    @classmethod
+    @DB.connection_context()
+    def delete_chunk_images(cls, doc, tenant_id):
+        page = 0
+        page_size = 1000
+        while True:
+            chunks = settings.docStoreConn.search(["img_id"], [], {"doc_id": doc.id}, [], OrderByExpr(),
+                                                  page * page_size, page_size, search.index_name(tenant_id),
+                                                  [doc.kb_id])
+            chunk_ids = settings.docStoreConn.get_doc_ids(chunks)
+            if not chunk_ids:
+                break
+            for cid in chunk_ids:
+                if settings.STORAGE_IMPL.obj_exist(doc.kb_id, cid):
+                    settings.STORAGE_IMPL.rm(doc.kb_id, cid)
+            page += 1
+
     @classmethod
     @DB.connection_context()
     def get_newly_uploaded(cls):
@@ -696,10 +703,14 @@ class DocumentService(CommonService):
             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)
+                if not isinstance(v, list):
+                    v = [v]
+                for vv in v:
+                    if vv not in meta[k]:
+                        if isinstance(vv, list) or isinstance(vv, dict):
+                            continue
+                        meta[k][vv] = []
+                    meta[k][vv].append(doc_id)
         return meta
 
     @classmethod
@@ -797,7 +808,10 @@ class DocumentService(CommonService):
                 match_provided = "match" in upd
                 if isinstance(meta[key], list):
                     if not match_provided:
-                        meta[key] = new_value
+                        if isinstance(new_value, list):
+                            meta[key] = dedupe_list(new_value)
+                        else:
+                            meta[key] = new_value
                         changed = True
                     else:
                         match_value = upd.get("match")
@@ -810,7 +824,7 @@ class DocumentService(CommonService):
                             else:
                                 new_list.append(item)
                         if replaced:
-                            meta[key] = new_list
+                            meta[key] = dedupe_list(new_list)
                             changed = True
                 else:
                     if not match_provided:
@@ -1230,8 +1244,8 @@ def doc_upload_and_parse(conversation_id, file_objs, user_id):
             d["q_%d_vec" % len(v)] = v
         for b in range(0, len(cks), es_bulk_size):
             if try_create_idx:
-                if not settings.docStoreConn.indexExist(idxnm, kb_id):
-                    settings.docStoreConn.createIdx(idxnm, kb_id, len(vectors[0]))
+                if not settings.docStoreConn.index_exist(idxnm, kb_id):
+                    settings.docStoreConn.create_idx(idxnm, kb_id, len(vectors[0]))
                 try_create_idx = False
             settings.docStoreConn.insert(cks[b:b + es_bulk_size], idxnm, kb_id)
 

api/db/services/evaluation_service.py

@@ -65,6 +65,7 @@ class EvaluationService(CommonService):
             (success, dataset_id or error_message)
         """
         try:
+            timestamp= current_timestamp()
             dataset_id = get_uuid()
             dataset = {
                 "id": dataset_id,
@@ -73,8 +74,8 @@ class EvaluationService(CommonService):
                 "description": description,
                 "kb_ids": kb_ids,
                 "created_by": user_id,
-                "create_time": current_timestamp(),
-                "update_time": current_timestamp(),
+                "create_time": timestamp,
+                "update_time": timestamp,
                 "status": StatusEnum.VALID.value
             }
 

api/db/services/knowledgebase_service.py

@@ -411,8 +411,6 @@ class KnowledgebaseService(CommonService):
         ok, _t = TenantService.get_by_id(tenant_id)
         if not ok:
             return False, get_data_error_result(message="Tenant not found.")
-        if kwargs.get("parser_config") and isinstance(kwargs["parser_config"], dict) and not kwargs["parser_config"].get("llm_id"):
-            kwargs["parser_config"]["llm_id"] = _t.llm_id
 
         # Build payload
         kb_id = get_uuid()
@@ -427,6 +425,7 @@ class KnowledgebaseService(CommonService):
 
         # Update parser_config (always override with validated default/merged config)
         payload["parser_config"] = get_parser_config(parser_id, kwargs.get("parser_config"))
+        payload["parser_config"]["llm_id"] = _t.llm_id
 
         return True, payload
 

api/db/services/langfuse_service.py

@@ -64,10 +64,13 @@ class TenantLangfuseService(CommonService):
 
     @classmethod
     def save(cls, **kwargs):
-        kwargs["create_time"] = current_timestamp()
-        kwargs["create_date"] = datetime_format(datetime.now())
-        kwargs["update_time"] = current_timestamp()
-        kwargs["update_date"] = datetime_format(datetime.now())
+        current_ts = current_timestamp()
+        current_date = datetime_format(datetime.now())
+
+        kwargs["create_time"] = current_ts
+        kwargs["create_date"] = current_date
+        kwargs["update_time"] = current_ts
+        kwargs["update_date"] = current_date
         obj = cls.model.create(**kwargs)
         return obj
 

api/db/services/memory_service.py

@@ -15,7 +15,6 @@
 #
 from typing import List
 
-from api.apps import current_user
 from api.db.db_models import DB, Memory, User
 from api.db.services import duplicate_name
 from api.db.services.common_service import CommonService
@@ -23,6 +22,7 @@ from api.utils.memory_utils import calculate_memory_type
 from api.constants import MEMORY_NAME_LIMIT
 from common.misc_utils import get_uuid
 from common.time_utils import get_format_time, current_timestamp
+from memory.utils.prompt_util import PromptAssembler
 
 
 class MemoryService(CommonService):
@@ -34,6 +34,17 @@ class MemoryService(CommonService):
     def get_by_memory_id(cls, memory_id: str):
         return cls.model.select().where(cls.model.id == memory_id).first()
 
+    @classmethod
+    @DB.connection_context()
+    def get_by_tenant_id(cls, tenant_id: str):
+        return cls.model.select().where(cls.model.tenant_id == tenant_id)
+
+    @classmethod
+    @DB.connection_context()
+    def get_all_memory(cls):
+        memory_list = cls.model.select()
+        return list(memory_list)
+
     @classmethod
     @DB.connection_context()
     def get_with_owner_name_by_id(cls, memory_id: str):
@@ -53,7 +64,9 @@ class MemoryService(CommonService):
             cls.model.forgetting_policy,
             cls.model.temperature,
             cls.model.system_prompt,
-            cls.model.user_prompt
+            cls.model.user_prompt,
+            cls.model.create_date,
+            cls.model.create_time
         ]
         memory = cls.model.select(*fields).join(User, on=(cls.model.tenant_id == User.id)).where(
             cls.model.id == memory_id
@@ -72,7 +85,9 @@ class MemoryService(CommonService):
             cls.model.memory_type,
             cls.model.storage_type,
             cls.model.permissions,
-            cls.model.description
+            cls.model.description,
+            cls.model.create_time,
+            cls.model.create_date
         ]
         memories = cls.model.select(*fields).join(User, on=(cls.model.tenant_id == User.id))
         if filter_dict.get("tenant_id"):
@@ -102,6 +117,8 @@ class MemoryService(CommonService):
         if len(memory_name) > MEMORY_NAME_LIMIT:
             return False, f"Memory name {memory_name} exceeds limit of {MEMORY_NAME_LIMIT}."
 
+        timestamp = current_timestamp()
+        format_time = get_format_time()
         # build create dict
         memory_info = {
             "id": get_uuid(),
@@ -110,10 +127,11 @@ class MemoryService(CommonService):
             "tenant_id": tenant_id,
             "embd_id": embd_id,
             "llm_id": llm_id,
-            "create_time": current_timestamp(),
-            "create_date": get_format_time(),
-            "update_time": current_timestamp(),
-            "update_date": get_format_time(),
+            "system_prompt": PromptAssembler.assemble_system_prompt({"memory_type": memory_type}),
+            "create_time": timestamp,
+            "create_date": format_time,
+            "update_time": timestamp,
+            "update_date": format_time,
         }
         obj = cls.model(**memory_info).save(force_insert=True)
 
@@ -126,16 +144,18 @@ class MemoryService(CommonService):
 
     @classmethod
     @DB.connection_context()
-    def update_memory(cls, memory_id: str, update_dict: dict):
+    def update_memory(cls, tenant_id: str, memory_id: str, update_dict: dict):
         if not update_dict:
             return 0
         if "temperature" in update_dict and isinstance(update_dict["temperature"], str):
             update_dict["temperature"] = float(update_dict["temperature"])
+        if "memory_type" in update_dict and isinstance(update_dict["memory_type"], list):
+            update_dict["memory_type"] = calculate_memory_type(update_dict["memory_type"])
         if "name" in update_dict:
             update_dict["name"] = duplicate_name(
                 cls.query,
                 name=update_dict["name"],
-                tenant_id=current_user.id
+                tenant_id=tenant_id
             )
         update_dict.update({
             "update_time": current_timestamp(),

api/db/services/pipeline_operation_log_service.py

@@ -169,11 +169,12 @@ class PipelineOperationLogService(CommonService):
             operation_status=operation_status,
             avatar=avatar,
         )
-        log["create_time"] = current_timestamp()
-        log["create_date"] = datetime_format(datetime.now())
-        log["update_time"] = current_timestamp()
-        log["update_date"] = datetime_format(datetime.now())
-
+        timestamp = current_timestamp()
+        datetime_now = datetime_format(datetime.now())
+        log["create_time"] = timestamp
+        log["create_date"] = datetime_now
+        log["update_time"] = timestamp
+        log["update_date"] = datetime_now
         with DB.atomic():
             obj = cls.save(**log)
 

api/db/services/search_service.py

@@ -28,10 +28,13 @@ class SearchService(CommonService):
 
     @classmethod
     def save(cls, **kwargs):
-        kwargs["create_time"] = current_timestamp()
-        kwargs["create_date"] = datetime_format(datetime.now())
-        kwargs["update_time"] = current_timestamp()
-        kwargs["update_date"] = datetime_format(datetime.now())
+        current_ts = current_timestamp()
+        current_date = datetime_format(datetime.now())
+
+        kwargs["create_time"] = current_ts
+        kwargs["create_date"] = current_date
+        kwargs["update_time"] = current_ts
+        kwargs["update_date"] = current_date
         obj = cls.model.create(**kwargs)
         return obj
 

api/db/services/user_service.py

@@ -116,10 +116,13 @@ class UserService(CommonService):
             kwargs["password"] = generate_password_hash(
                 str(kwargs["password"]))
 
-        kwargs["create_time"] = current_timestamp()
-        kwargs["create_date"] = datetime_format(datetime.now())
-        kwargs["update_time"] = current_timestamp()
-        kwargs["update_date"] = datetime_format(datetime.now())
+        current_ts = current_timestamp()
+        current_date = datetime_format(datetime.now())
+
+        kwargs["create_time"] = current_ts
+        kwargs["create_date"] = current_date
+        kwargs["update_time"] = current_ts
+        kwargs["update_date"] = current_date
         obj = cls.model(**kwargs).save(force_insert=True)
         return obj
 

api/utils/file_utils.py

@@ -42,7 +42,7 @@ def filename_type(filename):
     if re.match(r".*\.pdf$", filename):
         return FileType.PDF.value
 
-    if re.match(r".*\.(msg|eml|doc|docx|ppt|pptx|yml|xml|htm|json|jsonl|ldjson|csv|txt|ini|xls|xlsx|wps|rtf|hlp|pages|numbers|key|md|py|js|java|c|cpp|h|php|go|ts|sh|cs|kt|html|sql)$", filename):
+    if re.match(r".*\.(msg|eml|doc|docx|ppt|pptx|yml|xml|htm|json|jsonl|ldjson|csv|txt|ini|xls|xlsx|wps|rtf|hlp|pages|numbers|key|md|mdx|py|js|java|c|cpp|h|php|go|ts|sh|cs|kt|html|sql)$", filename):
         return FileType.DOC.value
 
     if re.match(r".*\.(wav|flac|ape|alac|wavpack|wv|mp3|aac|ogg|vorbis|opus)$", filename):

api/utils/web_utils.py

@@ -69,6 +69,7 @@ CONTENT_TYPE_MAP = {
     # Web
     "md": "text/markdown",
     "markdown": "text/markdown",
+    "mdx": "text/markdown",
     "htm": "text/html",
     "html": "text/html",
     "json": "application/json",

common/constants.py

@@ -128,14 +128,19 @@ class FileSource(StrEnum):
     R2 = "r2"
     OCI_STORAGE = "oci_storage"
     GOOGLE_CLOUD_STORAGE = "google_cloud_storage"
+    AIRTABLE = "airtable"
+    ASANA = "asana"
+    GITHUB = "github"
+    GITLAB = "gitlab"
+    IMAP = "imap"
 
-    
 class PipelineTaskType(StrEnum):
     PARSE = "Parse"
     DOWNLOAD = "Download"
     RAPTOR = "RAPTOR"
     GRAPH_RAG = "GraphRAG"
     MINDMAP = "Mindmap"
+    MEMORY = "Memory"
 
 
 VALID_PIPELINE_TASK_TYPES = {PipelineTaskType.PARSE, PipelineTaskType.DOWNLOAD, PipelineTaskType.RAPTOR,
@@ -170,7 +175,7 @@ class MemoryStorageType(StrEnum):
 
 
 class ForgettingPolicy(StrEnum):
-    FIFO = "fifo"
+    FIFO = "FIFO"
 
 
 # environment

common/data_source/__init__.py

@@ -36,6 +36,9 @@ from .sharepoint_connector import SharePointConnector
 from .teams_connector import TeamsConnector
 from .webdav_connector import WebDAVConnector
 from .moodle_connector import MoodleConnector
+from .airtable_connector import AirtableConnector
+from .asana_connector import AsanaConnector
+from .imap_connector import ImapConnector
 from .config import BlobType, DocumentSource
 from .models import Document, TextSection, ImageSection, BasicExpertInfo
 from .exceptions import (
@@ -70,5 +73,8 @@ __all__ = [
     "ConnectorValidationError",
     "CredentialExpiredError",
     "InsufficientPermissionsError",
-    "UnexpectedValidationError"
+    "UnexpectedValidationError",
+    "AirtableConnector",
+    "AsanaConnector",
+    "ImapConnector"
 ]

common/data_source/airtable_connector.py

@@ -0,0 +1,169 @@
+from datetime import datetime, timezone
+import logging
+from typing import Any, Generator
+
+import requests
+
+from pyairtable import Api as AirtableApi
+
+from common.data_source.config import AIRTABLE_CONNECTOR_SIZE_THRESHOLD, INDEX_BATCH_SIZE, DocumentSource
+from common.data_source.exceptions import ConnectorMissingCredentialError
+from common.data_source.interfaces import LoadConnector, PollConnector
+from common.data_source.models import Document, GenerateDocumentsOutput, SecondsSinceUnixEpoch
+from common.data_source.utils import extract_size_bytes, get_file_ext
+
+class AirtableClientNotSetUpError(PermissionError):
+    def __init__(self) -> None:
+        super().__init__(
+            "Airtable client is not set up. Did you forget to call load_credentials()?"
+        )
+
+
+class AirtableConnector(LoadConnector, PollConnector):
+    """
+    Lightweight Airtable connector.
+
+    This connector ingests Airtable attachments as raw blobs without
+    parsing file content or generating text/image sections.
+    """
+
+    def __init__(
+        self,
+        base_id: str,
+        table_name_or_id: str,
+        batch_size: int = INDEX_BATCH_SIZE,
+    ) -> None:
+        self.base_id = base_id
+        self.table_name_or_id = table_name_or_id
+        self.batch_size = batch_size
+        self._airtable_client: AirtableApi | None = None
+        self.size_threshold = AIRTABLE_CONNECTOR_SIZE_THRESHOLD
+
+    # -------------------------
+    # Credentials
+    # -------------------------
+    def load_credentials(self, credentials: dict[str, Any]) -> dict[str, Any] | None:
+        self._airtable_client = AirtableApi(credentials["airtable_access_token"])
+        return None
+
+    @property
+    def airtable_client(self) -> AirtableApi:
+        if not self._airtable_client:
+            raise AirtableClientNotSetUpError()
+        return self._airtable_client
+
+    # -------------------------
+    # Core logic
+    # -------------------------
+    def load_from_state(self) -> GenerateDocumentsOutput:
+        """
+        Fetch all Airtable records and ingest attachments as raw blobs.
+
+        Each attachment is converted into a single Document(blob=...).
+        """
+        if not self._airtable_client:
+            raise ConnectorMissingCredentialError("Airtable credentials not loaded")
+
+        table = self.airtable_client.table(self.base_id, self.table_name_or_id)
+        records = table.all()
+
+        logging.info(
+            f"Starting Airtable blob ingestion for table {self.table_name_or_id}, "
+            f"{len(records)} records found."
+        )
+
+        batch: list[Document] = []
+
+        for record in records:
+            print(record)
+            record_id = record.get("id")
+            fields = record.get("fields", {})
+            created_time = record.get("createdTime")
+
+            for field_value in fields.values():
+                # We only care about attachment fields (lists of dicts with url/filename)
+                if not isinstance(field_value, list):
+                    continue
+
+                for attachment in field_value:
+                    url = attachment.get("url")
+                    filename = attachment.get("filename")
+                    attachment_id = attachment.get("id")
+
+                    if not url or not filename or not attachment_id:
+                        continue
+
+                    try:
+                        resp = requests.get(url, timeout=30)
+                        resp.raise_for_status()
+                        content = resp.content
+                    except Exception:
+                        logging.exception(
+                            f"Failed to download attachment {filename} "
+                            f"(record={record_id})"
+                        )
+                        continue
+                    size_bytes = extract_size_bytes(attachment)
+                    if (
+                        self.size_threshold is not None
+                        and isinstance(size_bytes, int)
+                        and size_bytes > self.size_threshold
+                    ):
+                        logging.warning(
+                            f"{filename} exceeds size threshold of {self.size_threshold}. Skipping."
+                        )
+                        continue
+                    batch.append(
+                        Document(
+                            id=f"airtable:{record_id}:{attachment_id}",
+                            blob=content,
+                            source=DocumentSource.AIRTABLE,
+                            semantic_identifier=filename,
+                            extension=get_file_ext(filename),
+                            size_bytes=size_bytes if size_bytes else 0,
+                            doc_updated_at=datetime.strptime(created_time, "%Y-%m-%dT%H:%M:%S.%fZ").replace(tzinfo=timezone.utc)
+                        )
+                    )
+
+                    if len(batch) >= self.batch_size:
+                        yield batch
+                        batch = []
+
+        if batch:
+            yield batch
+
+    def poll_source(self, start: SecondsSinceUnixEpoch, end: SecondsSinceUnixEpoch) -> Generator[list[Document], None, None]:
+        """Poll source to get documents"""
+        start_dt = datetime.fromtimestamp(start, tz=timezone.utc)
+        end_dt = datetime.fromtimestamp(end, tz=timezone.utc)
+
+        for batch in self.load_from_state():
+            filtered: list[Document] = []
+
+            for doc in batch:
+                if not doc.doc_updated_at:
+                    continue
+
+                doc_dt = doc.doc_updated_at.astimezone(timezone.utc)
+
+                if start_dt <= doc_dt < end_dt:
+                    filtered.append(doc)
+
+            if filtered:
+                yield filtered
+
+if __name__ == "__main__":
+    import os
+
+    logging.basicConfig(level=logging.DEBUG)
+    connector = AirtableConnector("xxx","xxx")
+    connector.load_credentials({"airtable_access_token": os.environ.get("AIRTABLE_ACCESS_TOKEN")})
+    connector.validate_connector_settings()
+    document_batches = connector.load_from_state()
+    try:
+        first_batch = next(document_batches)
+        print(f"Loaded {len(first_batch)} documents in first batch.")
+        for doc in first_batch:
+            print(f"- {doc.semantic_identifier} ({doc.size_bytes} bytes)")
+    except StopIteration:
+        print("No documents available in Dropbox.")

common/data_source/asana_connector.py

@@ -0,0 +1,454 @@
+from collections.abc import Iterator
+import time
+from datetime import datetime
+import logging
+from typing import Any, Dict
+import asana
+import requests
+from common.data_source.config import CONTINUE_ON_CONNECTOR_FAILURE, INDEX_BATCH_SIZE, DocumentSource
+from common.data_source.interfaces import LoadConnector, PollConnector
+from common.data_source.models import Document, GenerateDocumentsOutput, SecondsSinceUnixEpoch
+from common.data_source.utils import extract_size_bytes, get_file_ext
+
+
+
+# https://github.com/Asana/python-asana/tree/master?tab=readme-ov-file#documentation-for-api-endpoints
+class AsanaTask:
+    def __init__(
+        self,
+        id: str,
+        title: str,
+        text: str,
+        link: str,
+        last_modified: datetime,
+        project_gid: str,
+        project_name: str,
+    ) -> None:
+        self.id = id
+        self.title = title
+        self.text = text
+        self.link = link
+        self.last_modified = last_modified
+        self.project_gid = project_gid
+        self.project_name = project_name
+
+    def __str__(self) -> str:
+        return f"ID: {self.id}\nTitle: {self.title}\nLast modified: {self.last_modified}\nText: {self.text}"
+
+
+class AsanaAPI:
+    def __init__(
+        self, api_token: str, workspace_gid: str, team_gid: str | None
+    ) -> None:
+        self._user = None
+        self.workspace_gid = workspace_gid
+        self.team_gid = team_gid
+
+        self.configuration = asana.Configuration()
+        self.api_client = asana.ApiClient(self.configuration)
+        self.tasks_api = asana.TasksApi(self.api_client)
+        self.attachments_api = asana.AttachmentsApi(self.api_client)
+        self.stories_api = asana.StoriesApi(self.api_client)
+        self.users_api = asana.UsersApi(self.api_client)
+        self.project_api = asana.ProjectsApi(self.api_client)
+        self.project_memberships_api = asana.ProjectMembershipsApi(self.api_client)
+        self.workspaces_api = asana.WorkspacesApi(self.api_client)
+
+        self.api_error_count = 0
+        self.configuration.access_token = api_token
+        self.task_count = 0
+
+    def get_tasks(
+        self, project_gids: list[str] | None, start_date: str
+    ) -> Iterator[AsanaTask]:
+        """Get all tasks from the projects with the given gids that were modified since the given date.
+        If project_gids is None, get all tasks from all projects in the workspace."""
+        logging.info("Starting to fetch Asana projects")
+        projects = self.project_api.get_projects(
+            opts={
+                "workspace": self.workspace_gid,
+                "opt_fields": "gid,name,archived,modified_at",
+            }
+        )
+        start_seconds = int(time.mktime(datetime.now().timetuple()))
+        projects_list = []
+        project_count = 0
+        for project_info in projects:
+            project_gid = project_info["gid"]
+            if project_gids is None or project_gid in project_gids:
+                projects_list.append(project_gid)
+            else:
+                logging.debug(
+                    f"Skipping project: {project_gid} - not in accepted project_gids"
+                )
+            project_count += 1
+            if project_count % 100 == 0:
+                logging.info(f"Processed {project_count} projects")
+        logging.info(f"Found {len(projects_list)} projects to process")
+        for project_gid in projects_list:
+            for task in self._get_tasks_for_project(
+                project_gid, start_date, start_seconds
+            ):
+                yield task
+        logging.info(f"Completed fetching {self.task_count} tasks from Asana")
+        if self.api_error_count > 0:
+            logging.warning(
+                f"Encountered {self.api_error_count} API errors during task fetching"
+            )
+
+    def _get_tasks_for_project(
+        self, project_gid: str, start_date: str, start_seconds: int
+    ) -> Iterator[AsanaTask]:
+        project = self.project_api.get_project(project_gid, opts={})
+        project_name = project.get("name", project_gid)
+        team = project.get("team") or {}
+        team_gid = team.get("gid")
+
+        if project.get("archived"):
+            logging.info(f"Skipping archived project: {project_name} ({project_gid})")
+            return
+        if not team_gid:
+            logging.info(
+                f"Skipping project without a team: {project_name} ({project_gid})"
+            )
+            return
+        if project.get("privacy_setting") == "private":
+            if self.team_gid and team_gid != self.team_gid:
+                logging.info(
+                    f"Skipping private project not in configured team: {project_name} ({project_gid})"
+                )
+                return
+            logging.info(
+                f"Processing private project in configured team: {project_name} ({project_gid})"
+            )
+
+        simple_start_date = start_date.split(".")[0].split("+")[0]
+        logging.info(
+            f"Fetching tasks modified since {simple_start_date} for project: {project_name} ({project_gid})"
+        )
+
+        opts = {
+            "opt_fields": "name,memberships,memberships.project,completed_at,completed_by,created_at,"
+            "created_by,custom_fields,dependencies,due_at,due_on,external,html_notes,liked,likes,"
+            "modified_at,notes,num_hearts,parent,projects,resource_subtype,resource_type,start_on,"
+            "workspace,permalink_url",
+            "modified_since": start_date,
+        }
+        tasks_from_api = self.tasks_api.get_tasks_for_project(project_gid, opts)
+        for data in tasks_from_api:
+            self.task_count += 1
+            if self.task_count % 10 == 0:
+                end_seconds = time.mktime(datetime.now().timetuple())
+                runtime_seconds = end_seconds - start_seconds
+                if runtime_seconds > 0:
+                    logging.info(
+                        f"Processed {self.task_count} tasks in {runtime_seconds:.0f} seconds "
+                        f"({self.task_count / runtime_seconds:.2f} tasks/second)"
+                    )
+
+            logging.debug(f"Processing Asana task: {data['name']}")
+
+            text = self._construct_task_text(data)
+
+            try:
+                text += self._fetch_and_add_comments(data["gid"])
+
+                last_modified_date = self.format_date(data["modified_at"])
+                text += f"Last modified: {last_modified_date}\n"
+
+                task = AsanaTask(
+                    id=data["gid"],
+                    title=data["name"],
+                    text=text,
+                    link=data["permalink_url"],
+                    last_modified=datetime.fromisoformat(data["modified_at"]),
+                    project_gid=project_gid,
+                    project_name=project_name,
+                )
+                yield task
+            except Exception:
+                logging.error(
+                    f"Error processing task {data['gid']} in project {project_gid}",
+                    exc_info=True,
+                )
+                self.api_error_count += 1
+
+    def _construct_task_text(self, data: Dict) -> str:
+        text = f"{data['name']}\n\n"
+
+        if data["notes"]:
+            text += f"{data['notes']}\n\n"
+
+        if data["created_by"] and data["created_by"]["gid"]:
+            creator = self.get_user(data["created_by"]["gid"])["name"]
+            created_date = self.format_date(data["created_at"])
+            text += f"Created by: {creator} on {created_date}\n"
+
+        if data["due_on"]:
+            due_date = self.format_date(data["due_on"])
+            text += f"Due date: {due_date}\n"
+
+        if data["completed_at"]:
+            completed_date = self.format_date(data["completed_at"])
+            text += f"Completed on: {completed_date}\n"
+
+        text += "\n"
+        return text
+
+    def _fetch_and_add_comments(self, task_gid: str) -> str:
+        text = ""
+        stories_opts: Dict[str, str] = {}
+        story_start = time.time()
+        stories = self.stories_api.get_stories_for_task(task_gid, stories_opts)
+
+        story_count = 0
+        comment_count = 0
+
+        for story in stories:
+            story_count += 1
+            if story["resource_subtype"] == "comment_added":
+                comment = self.stories_api.get_story(
+                    story["gid"], opts={"opt_fields": "text,created_by,created_at"}
+                )
+                commenter = self.get_user(comment["created_by"]["gid"])["name"]
+                text += f"Comment by {commenter}: {comment['text']}\n\n"
+                comment_count += 1
+
+        story_duration = time.time() - story_start
+        logging.debug(
+            f"Processed {story_count} stories (including {comment_count} comments) in {story_duration:.2f} seconds"
+        )
+
+        return text
+
+    def get_attachments(self, task_gid: str) -> list[dict]:
+        """
+        Fetch full attachment info (including download_url) for a task.
+        """
+        attachments: list[dict] = []
+
+        try:
+            # Step 1: list attachment compact records
+            for att in self.attachments_api.get_attachments_for_object(
+                parent=task_gid,
+                opts={}
+            ):
+                gid = att.get("gid")
+                if not gid:
+                    continue
+
+                try:
+                    # Step 2: expand to full attachment
+                    full = self.attachments_api.get_attachment(
+                        attachment_gid=gid,
+                        opts={
+                            "opt_fields": "name,download_url,size,created_at"
+                        }
+                    )
+
+                    if full.get("download_url"):
+                        attachments.append(full)
+
+                except Exception:
+                    logging.exception(
+                        f"Failed to fetch attachment detail {gid} for task {task_gid}"
+                    )
+                    self.api_error_count += 1
+
+        except Exception:
+            logging.exception(f"Failed to list attachments for task {task_gid}")
+            self.api_error_count += 1
+
+        return attachments
+
+    def get_accessible_emails(
+        self,
+        workspace_id: str,
+        project_ids: list[str] | None,
+        team_id: str | None,
+    ):
+
+        ws_users = self.users_api.get_users(
+            opts={
+                "workspace": workspace_id,
+                "opt_fields": "gid,name,email"
+            }
+        )
+
+        workspace_users = {
+            u["gid"]: u.get("email")
+            for u in ws_users
+            if u.get("email")
+        }
+
+        if not project_ids:
+            return set(workspace_users.values())
+
+
+        project_emails = set()
+
+        for pid in project_ids:
+            project = self.project_api.get_project(
+                pid,
+                opts={"opt_fields": "team,privacy_setting"}
+            )
+
+            if project["privacy_setting"] == "private":
+                if team_id and project.get("team", {}).get("gid") != team_id:
+                    continue
+
+            memberships = self.project_memberships_api.get_project_membership(
+                pid,
+                opts={"opt_fields": "user.gid,user.email"}
+            )
+
+            for m in memberships:
+                email = m["user"].get("email")
+                if email:
+                    project_emails.add(email)
+
+        return project_emails
+
+    def get_user(self, user_gid: str) -> Dict:
+        if self._user is not None:
+            return self._user
+        self._user = self.users_api.get_user(user_gid, {"opt_fields": "name,email"})
+
+        if not self._user:
+            logging.warning(f"Unable to fetch user information for user_gid: {user_gid}")
+            return {"name": "Unknown"}
+        return self._user
+
+    def format_date(self, date_str: str) -> str:
+        date = datetime.fromisoformat(date_str)
+        return time.strftime("%Y-%m-%d", date.timetuple())
+
+    def get_time(self) -> str:
+        return time.strftime("%Y-%m-%d %H:%M:%S", time.localtime())
+
+
+class AsanaConnector(LoadConnector, PollConnector):
+    def __init__(
+        self,
+        asana_workspace_id: str,
+        asana_project_ids: str | None = None,
+        asana_team_id: str | None = None,
+        batch_size: int = INDEX_BATCH_SIZE,
+        continue_on_failure: bool = CONTINUE_ON_CONNECTOR_FAILURE,
+    ) -> None:
+        self.workspace_id = asana_workspace_id
+        self.project_ids_to_index: list[str] | None = (
+            asana_project_ids.split(",") if asana_project_ids else None
+        )
+        self.asana_team_id = asana_team_id if asana_team_id else None
+        self.batch_size = batch_size
+        self.continue_on_failure = continue_on_failure
+        self.size_threshold = None
+        logging.info(
+            f"AsanaConnector initialized with workspace_id: {asana_workspace_id}"
+        )
+
+    def load_credentials(self, credentials: dict[str, Any]) -> dict[str, Any] | None:
+        self.api_token = credentials["asana_api_token_secret"]
+        self.asana_client = AsanaAPI(
+            api_token=self.api_token,
+            workspace_gid=self.workspace_id,
+            team_gid=self.asana_team_id,
+        )
+        self.workspace_users_email = self.asana_client.get_accessible_emails(self.workspace_id, self.project_ids_to_index, self.asana_team_id)
+        logging.info("Asana credentials loaded and API client initialized")
+        return None
+
+    def poll_source(
+        self, start: SecondsSinceUnixEpoch, end: SecondsSinceUnixEpoch | None
+    ) -> GenerateDocumentsOutput:
+        start_time = datetime.fromtimestamp(start).isoformat()
+        logging.info(f"Starting Asana poll from {start_time}")
+        docs_batch: list[Document] = []
+        tasks = self.asana_client.get_tasks(self.project_ids_to_index, start_time)
+        for task in tasks:
+            docs = self._task_to_documents(task)
+            docs_batch.extend(docs)
+
+            if len(docs_batch) >= self.batch_size:
+                logging.info(f"Yielding batch of {len(docs_batch)} documents")
+                yield docs_batch
+                docs_batch = []
+
+        if docs_batch:
+            logging.info(f"Yielding final batch of {len(docs_batch)} documents")
+            yield docs_batch
+
+        logging.info("Asana poll completed")
+
+    def load_from_state(self) -> GenerateDocumentsOutput:
+        logging.info("Starting full index of all Asana tasks")
+        return self.poll_source(start=0, end=None)
+
+    def _task_to_documents(self, task: AsanaTask) -> list[Document]:
+        docs: list[Document] = []
+
+        attachments = self.asana_client.get_attachments(task.id)
+
+        for att in attachments:
+            try:
+                resp = requests.get(att["download_url"], timeout=30)
+                resp.raise_for_status()
+                file_blob = resp.content
+                filename = att.get("name", "attachment")
+                size_bytes = extract_size_bytes(att)
+                if (
+                    self.size_threshold is not None
+                    and isinstance(size_bytes, int)
+                    and size_bytes > self.size_threshold
+                ):
+                    logging.warning(
+                        f"{filename} exceeds size threshold of {self.size_threshold}. Skipping."
+                    )
+                    continue
+                docs.append(
+                    Document(
+                        id=f"asana:{task.id}:{att['gid']}",
+                        blob=file_blob,
+                        extension=get_file_ext(filename) or "",
+                        size_bytes=size_bytes,
+                        doc_updated_at=task.last_modified,
+                        source=DocumentSource.ASANA,
+                        semantic_identifier=filename,
+                        primary_owners=list(self.workspace_users_email),
+                    )
+                )
+            except Exception:
+                logging.exception(
+                    f"Failed to download attachment {att.get('gid')} for task {task.id}"
+                )
+
+        return docs
+
+
+
+if __name__ == "__main__":
+    import time
+    import os
+
+    logging.info("Starting Asana connector test")
+    connector = AsanaConnector(
+        os.environ["WORKSPACE_ID"],
+        os.environ["PROJECT_IDS"],
+        os.environ["TEAM_ID"],
+    )
+    connector.load_credentials(
+        {
+            "asana_api_token_secret": os.environ["API_TOKEN"],
+        }
+    )
+    logging.info("Loading all documents from Asana")
+    all_docs = connector.load_from_state()
+    current = time.time()
+    one_day_ago = current - 24 * 60 * 60  # 1 day
+    logging.info("Polling for documents updated in the last 24 hours")
+    latest_docs = connector.poll_source(one_day_ago, current)
+    for docs in all_docs:
+        for doc in docs:
+            print(doc.id)
+    logging.info("Asana connector test completed")

common/data_source/config.py

@@ -53,7 +53,13 @@ class DocumentSource(str, Enum):
     S3_COMPATIBLE = "s3_compatible"
     DROPBOX = "dropbox"
     BOX = "box"
+    AIRTABLE = "airtable"
+    ASANA = "asana"
+    GITHUB = "github"
+    GITLAB = "gitlab"
+    IMAP = "imap"
 
+    
 class FileOrigin(str, Enum):
     """File origins"""
     CONNECTOR = "connector"
@@ -230,6 +236,8 @@ _REPLACEMENT_EXPANSIONS = "body.view.value"
 
 BOX_WEB_OAUTH_REDIRECT_URI = os.environ.get("BOX_WEB_OAUTH_REDIRECT_URI", "http://localhost:9380/v1/connector/box/oauth/web/callback")
 
+GITHUB_CONNECTOR_BASE_URL = os.environ.get("GITHUB_CONNECTOR_BASE_URL") or None
+
 class HtmlBasedConnectorTransformLinksStrategy(str, Enum):
     # remove links entirely
     STRIP = "strip"
@@ -251,6 +259,18 @@ WEB_CONNECTOR_IGNORED_ELEMENTS = os.environ.get(
     "WEB_CONNECTOR_IGNORED_ELEMENTS", "nav,footer,meta,script,style,symbol,aside"
 ).split(",")
 
+AIRTABLE_CONNECTOR_SIZE_THRESHOLD = int(
+    os.environ.get("AIRTABLE_CONNECTOR_SIZE_THRESHOLD", 10 * 1024 * 1024)
+)
+
+ASANA_CONNECTOR_SIZE_THRESHOLD = int(
+    os.environ.get("ASANA_CONNECTOR_SIZE_THRESHOLD", 10 * 1024 * 1024)
+)
+
+IMAP_CONNECTOR_SIZE_THRESHOLD = int(
+    os.environ.get("IMAP_CONNECTOR_SIZE_THRESHOLD", 10 * 1024 * 1024)
+)
+
 _USER_NOT_FOUND = "Unknown Confluence User"
 
 _COMMENT_EXPANSION_FIELDS = ["body.storage.value"]

common/data_source/connector_runner.py

@@ -0,0 +1,217 @@
+import sys
+import time
+import logging
+from collections.abc import Generator
+from datetime import datetime
+from typing import Generic
+from typing import TypeVar
+from common.data_source.interfaces import (
+    BaseConnector,
+    CheckpointedConnector,
+    CheckpointedConnectorWithPermSync,
+    CheckpointOutput,
+    LoadConnector,
+    PollConnector,
+)
+from common.data_source.models import ConnectorCheckpoint, ConnectorFailure, Document
+
+
+TimeRange = tuple[datetime, datetime]
+
+CT = TypeVar("CT", bound=ConnectorCheckpoint)
+
+
+def batched_doc_ids(
+    checkpoint_connector_generator: CheckpointOutput[CT],
+    batch_size: int,
+) -> Generator[set[str], None, None]:
+    batch: set[str] = set()
+    for document, failure, next_checkpoint in CheckpointOutputWrapper[CT]()(
+        checkpoint_connector_generator
+    ):
+        if document is not None:
+            batch.add(document.id)
+        elif (
+            failure and failure.failed_document and failure.failed_document.document_id
+        ):
+            batch.add(failure.failed_document.document_id)
+
+        if len(batch) >= batch_size:
+            yield batch
+            batch = set()
+    if len(batch) > 0:
+        yield batch
+
+
+class CheckpointOutputWrapper(Generic[CT]):
+    """
+    Wraps a CheckpointOutput generator to give things back in a more digestible format,
+    specifically for Document outputs.
+    The connector format is easier for the connector implementor (e.g. it enforces exactly
+    one new checkpoint is returned AND that the checkpoint is at the end), thus the different
+    formats.
+    """
+
+    def __init__(self) -> None:
+        self.next_checkpoint: CT | None = None
+
+    def __call__(
+        self,
+        checkpoint_connector_generator: CheckpointOutput[CT],
+    ) -> Generator[
+        tuple[Document | None, ConnectorFailure | None, CT | None],
+        None,
+        None,
+    ]:
+        # grabs the final return value and stores it in the `next_checkpoint` variable
+        def _inner_wrapper(
+            checkpoint_connector_generator: CheckpointOutput[CT],
+        ) -> CheckpointOutput[CT]:
+            self.next_checkpoint = yield from checkpoint_connector_generator
+            return self.next_checkpoint  # not used
+
+        for document_or_failure in _inner_wrapper(checkpoint_connector_generator):
+            if isinstance(document_or_failure, Document):
+                yield document_or_failure, None, None
+            elif isinstance(document_or_failure, ConnectorFailure):
+                yield None, document_or_failure, None
+            else:
+                raise ValueError(
+                    f"Invalid document_or_failure type: {type(document_or_failure)}"
+                )
+
+        if self.next_checkpoint is None:
+            raise RuntimeError(
+                "Checkpoint is None. This should never happen - the connector should always return a checkpoint."
+            )
+
+        yield None, None, self.next_checkpoint
+
+
+class ConnectorRunner(Generic[CT]):
+    """
+    Handles:
+        - Batching
+        - Additional exception logging
+        - Combining different connector types to a single interface
+    """
+
+    def __init__(
+        self,
+        connector: BaseConnector,
+        batch_size: int,
+        # cannot be True for non-checkpointed connectors
+        include_permissions: bool,
+        time_range: TimeRange | None = None,
+    ):
+        if not isinstance(connector, CheckpointedConnector) and include_permissions:
+            raise ValueError(
+                "include_permissions cannot be True for non-checkpointed connectors"
+            )
+
+        self.connector = connector
+        self.time_range = time_range
+        self.batch_size = batch_size
+        self.include_permissions = include_permissions
+
+        self.doc_batch: list[Document] = []
+
+    def run(self, checkpoint: CT) -> Generator[
+        tuple[list[Document] | None, ConnectorFailure | None, CT | None],
+        None,
+        None,
+    ]:
+        """Adds additional exception logging to the connector."""
+        try:
+            if isinstance(self.connector, CheckpointedConnector):
+                if self.time_range is None:
+                    raise ValueError("time_range is required for CheckpointedConnector")
+
+                start = time.monotonic()
+                if self.include_permissions:
+                    if not isinstance(
+                        self.connector, CheckpointedConnectorWithPermSync
+                    ):
+                        raise ValueError(
+                            "Connector does not support permission syncing"
+                        )
+                    load_from_checkpoint = (
+                        self.connector.load_from_checkpoint_with_perm_sync
+                    )
+                else:
+                    load_from_checkpoint = self.connector.load_from_checkpoint
+                checkpoint_connector_generator = load_from_checkpoint(
+                    start=self.time_range[0].timestamp(),
+                    end=self.time_range[1].timestamp(),
+                    checkpoint=checkpoint,
+                )
+                next_checkpoint: CT | None = None
+                # this is guaranteed to always run at least once with next_checkpoint being non-None
+                for document, failure, next_checkpoint in CheckpointOutputWrapper[CT]()(
+                    checkpoint_connector_generator
+                ):
+                    if document is not None and isinstance(document, Document):
+                        self.doc_batch.append(document)
+
+                    if failure is not None:
+                        yield None, failure, None
+
+                    if len(self.doc_batch) >= self.batch_size:
+                        yield self.doc_batch, None, None
+                        self.doc_batch = []
+
+                # yield remaining documents
+                if len(self.doc_batch) > 0:
+                    yield self.doc_batch, None, None
+                    self.doc_batch = []
+
+                yield None, None, next_checkpoint
+
+                logging.debug(
+                    f"Connector took {time.monotonic() - start} seconds to get to the next checkpoint."
+                )
+
+            else:
+                finished_checkpoint = self.connector.build_dummy_checkpoint()
+                finished_checkpoint.has_more = False
+
+                if isinstance(self.connector, PollConnector):
+                    if self.time_range is None:
+                        raise ValueError("time_range is required for PollConnector")
+
+                    for document_batch in self.connector.poll_source(
+                        start=self.time_range[0].timestamp(),
+                        end=self.time_range[1].timestamp(),
+                    ):
+                        yield document_batch, None, None
+
+                    yield None, None, finished_checkpoint
+                elif isinstance(self.connector, LoadConnector):
+                    for document_batch in self.connector.load_from_state():
+                        yield document_batch, None, None
+
+                    yield None, None, finished_checkpoint
+                else:
+                    raise ValueError(f"Invalid connector. type: {type(self.connector)}")
+        except Exception:
+            exc_type, _, exc_traceback = sys.exc_info()
+
+            # Traverse the traceback to find the last frame where the exception was raised
+            tb = exc_traceback
+            if tb is None:
+                logging.error("No traceback found for exception")
+                raise
+
+            while tb.tb_next:
+                tb = tb.tb_next  # Move to the next frame in the traceback
+
+            # Get the local variables from the frame where the exception occurred
+            local_vars = tb.tb_frame.f_locals
+            local_vars_str = "\n".join(
+                f"{key}: {value}" for key, value in local_vars.items()
+            )
+            logging.error(
+                f"Error in connector. type: {exc_type};\n"
+                f"local_vars below -> \n{local_vars_str[:1024]}"
+            )
+            raise

common/data_source/file_types.py

@@ -18,6 +18,7 @@ class UploadMimeTypes:
         "text/plain",
         "text/markdown",
         "text/x-markdown",
+        "text/mdx",
         "text/x-config",
         "text/tab-separated-values",
         "application/json",

common/data_source/github/connector.py

@@ -0,0 +1,973 @@
+import copy
+import logging
+from collections.abc import Callable
+from collections.abc import Generator
+from datetime import datetime
+from datetime import timedelta
+from datetime import timezone
+from enum import Enum
+from typing import Any
+from typing import cast
+
+from github import Github, Auth
+from github import RateLimitExceededException
+from github import Repository
+from github.GithubException import GithubException
+from github.Issue import Issue
+from github.NamedUser import NamedUser
+from github.PaginatedList import PaginatedList
+from github.PullRequest import PullRequest
+from pydantic import BaseModel
+from typing_extensions import override
+from common.data_source.google_util.util import sanitize_filename
+from common.data_source.config import DocumentSource, GITHUB_CONNECTOR_BASE_URL
+from common.data_source.exceptions import (
+    ConnectorMissingCredentialError,
+    ConnectorValidationError,
+    CredentialExpiredError,
+    InsufficientPermissionsError,
+    UnexpectedValidationError,
+)
+from common.data_source.interfaces import CheckpointedConnectorWithPermSyncGH, CheckpointOutput
+from common.data_source.models import (
+    ConnectorCheckpoint,
+    ConnectorFailure,
+    Document,
+    DocumentFailure,
+    ExternalAccess,
+    SecondsSinceUnixEpoch,
+)
+from common.data_source.connector_runner import ConnectorRunner
+from .models import SerializedRepository
+from .rate_limit_utils import sleep_after_rate_limit_exception
+from .utils import deserialize_repository
+from .utils import get_external_access_permission
+
+ITEMS_PER_PAGE = 100
+CURSOR_LOG_FREQUENCY = 50
+
+_MAX_NUM_RATE_LIMIT_RETRIES = 5
+
+ONE_DAY = timedelta(days=1)
+SLIM_BATCH_SIZE = 100
+# Cases
+# X (from start) standard run, no fallback to cursor-based pagination
+# X (from start) standard run errors, fallback to cursor-based pagination
+#  X error in the middle of a page
+#  X no errors: run to completion
+# X (from checkpoint) standard run, no fallback to cursor-based pagination
+# X (from checkpoint) continue from cursor-based pagination
+#  - retrying
+#  - no retrying
+
+# things to check:
+# checkpoint state on return
+# checkpoint progress (no infinite loop)
+
+
+class DocMetadata(BaseModel):
+    repo: str
+
+
+def get_nextUrl_key(pag_list: PaginatedList[PullRequest | Issue]) -> str:
+    if "_PaginatedList__nextUrl" in pag_list.__dict__:
+        return "_PaginatedList__nextUrl"
+    for key in pag_list.__dict__:
+        if "__nextUrl" in key:
+            return key
+    for key in pag_list.__dict__:
+        if "nextUrl" in key:
+            return key
+    return ""
+
+
+def get_nextUrl(
+    pag_list: PaginatedList[PullRequest | Issue], nextUrl_key: str
+) -> str | None:
+    return getattr(pag_list, nextUrl_key) if nextUrl_key else None
+
+
+def set_nextUrl(
+    pag_list: PaginatedList[PullRequest | Issue], nextUrl_key: str, nextUrl: str
+) -> None:
+    if nextUrl_key:
+        setattr(pag_list, nextUrl_key, nextUrl)
+    elif nextUrl:
+        raise ValueError("Next URL key not found: " + str(pag_list.__dict__))
+
+
+def _paginate_until_error(
+    git_objs: Callable[[], PaginatedList[PullRequest | Issue]],
+    cursor_url: str | None,
+    prev_num_objs: int,
+    cursor_url_callback: Callable[[str | None, int], None],
+    retrying: bool = False,
+) -> Generator[PullRequest | Issue, None, None]:
+    num_objs = prev_num_objs
+    pag_list = git_objs()
+    nextUrl_key = get_nextUrl_key(pag_list)
+    if cursor_url:
+        set_nextUrl(pag_list, nextUrl_key, cursor_url)
+    elif retrying:
+        # if we are retrying, we want to skip the objects retrieved
+        # over previous calls. Unfortunately, this WILL retrieve all
+        # pages before the one we are resuming from, so we really
+        # don't want this case to be hit often
+        logging.warning(
+            "Retrying from a previous cursor-based pagination call. "
+            "This will retrieve all pages before the one we are resuming from, "
+            "which may take a while and consume many API calls."
+        )
+        pag_list = cast(PaginatedList[PullRequest | Issue], pag_list[prev_num_objs:])
+        num_objs = 0
+
+    try:
+        # this for loop handles cursor-based pagination
+        for issue_or_pr in pag_list:
+            num_objs += 1
+            yield issue_or_pr
+            # used to store the current cursor url in the checkpoint. This value
+            # is updated during iteration over pag_list.
+            cursor_url_callback(get_nextUrl(pag_list, nextUrl_key), num_objs)
+
+            if num_objs % CURSOR_LOG_FREQUENCY == 0:
+                logging.info(
+                    f"Retrieved {num_objs} objects with current cursor url: {get_nextUrl(pag_list, nextUrl_key)}"
+                )
+
+    except Exception as e:
+        logging.exception(f"Error during cursor-based pagination: {e}")
+        if num_objs - prev_num_objs > 0:
+            raise
+
+        if get_nextUrl(pag_list, nextUrl_key) is not None and not retrying:
+            logging.info(
+                "Assuming that this error is due to cursor "
+                "expiration because no objects were retrieved. "
+                "Retrying from the first page."
+            )
+            yield from _paginate_until_error(
+                git_objs, None, prev_num_objs, cursor_url_callback, retrying=True
+            )
+            return
+
+        # for no cursor url or if we reach this point after a retry, raise the error
+        raise
+
+
+def _get_batch_rate_limited(
+    # We pass in a callable because we want git_objs to produce a fresh
+    # PaginatedList each time it's called to avoid using the same object for cursor-based pagination
+    # from a partial offset-based pagination call.
+    git_objs: Callable[[], PaginatedList],
+    page_num: int,
+    cursor_url: str | None,
+    prev_num_objs: int,
+    cursor_url_callback: Callable[[str | None, int], None],
+    github_client: Github,
+    attempt_num: int = 0,
+) -> Generator[PullRequest | Issue, None, None]:
+    if attempt_num > _MAX_NUM_RATE_LIMIT_RETRIES:
+        raise RuntimeError(
+            "Re-tried fetching batch too many times. Something is going wrong with fetching objects from Github"
+        )
+    try:
+        if cursor_url:
+            # when this is set, we are resuming from an earlier
+            # cursor-based pagination call.
+            yield from _paginate_until_error(
+                git_objs, cursor_url, prev_num_objs, cursor_url_callback
+            )
+            return
+        objs = list(git_objs().get_page(page_num))
+        # fetch all data here to disable lazy loading later
+        # this is needed to capture the rate limit exception here (if one occurs)
+        for obj in objs:
+            if hasattr(obj, "raw_data"):
+                getattr(obj, "raw_data")
+        yield from objs
+    except RateLimitExceededException:
+        sleep_after_rate_limit_exception(github_client)
+        yield from _get_batch_rate_limited(
+            git_objs,
+            page_num,
+            cursor_url,
+            prev_num_objs,
+            cursor_url_callback,
+            github_client,
+            attempt_num + 1,
+        )
+    except GithubException as e:
+        if not (
+            e.status == 422
+            and (
+                "cursor" in (e.message or "")
+                or "cursor" in (e.data or {}).get("message", "")
+            )
+        ):
+            raise
+        # Fallback to a cursor-based pagination strategy
+        # This can happen for "large datasets," but there's no documentation
+        # On the error on the web as far as we can tell.
+        # Error message:
+        # "Pagination with the page parameter is not supported for large datasets,
+        # please use cursor based pagination (after/before)"
+        yield from _paginate_until_error(
+            git_objs, cursor_url, prev_num_objs, cursor_url_callback
+        )
+
+
+def _get_userinfo(user: NamedUser) -> dict[str, str]:
+    def _safe_get(attr_name: str) -> str | None:
+        try:
+            return cast(str | None, getattr(user, attr_name))
+        except GithubException:
+            logging.debug(f"Error getting {attr_name} for user")
+            return None
+
+    return {
+        k: v
+        for k, v in {
+            "login": _safe_get("login"),
+            "name": _safe_get("name"),
+            "email": _safe_get("email"),
+        }.items()
+        if v is not None
+    }
+
+
+def _convert_pr_to_document(
+    pull_request: PullRequest, repo_external_access: ExternalAccess | None
+) -> Document:
+    repo_name = pull_request.base.repo.full_name if pull_request.base else ""
+    doc_metadata = DocMetadata(repo=repo_name)
+    file_content_byte = pull_request.body.encode('utf-8') if pull_request.body else b""
+    name = sanitize_filename(pull_request.title, "md")
+
+    return Document(
+        id=pull_request.html_url,
+        blob= file_content_byte,
+        source=DocumentSource.GITHUB,
+        external_access=repo_external_access,
+        semantic_identifier=f"{pull_request.number}:{name}",
+        # updated_at is UTC time but is timezone unaware, explicitly add UTC
+        # as there is logic in indexing to prevent wrong timestamped docs
+        # due to local time discrepancies with UTC
+        doc_updated_at=(
+            pull_request.updated_at.replace(tzinfo=timezone.utc)
+            if pull_request.updated_at
+            else None
+        ),
+        extension=".md",
+        # this metadata is used in perm sync
+        size_bytes=len(file_content_byte) if file_content_byte else 0,
+        primary_owners=[],
+        doc_metadata=doc_metadata.model_dump(),
+        metadata={
+            k: [str(vi) for vi in v] if isinstance(v, list) else str(v)
+            for k, v in {
+                "object_type": "PullRequest",
+                "id": pull_request.number,
+                "merged": pull_request.merged,
+                "state": pull_request.state,
+                "user": _get_userinfo(pull_request.user) if pull_request.user else None,
+                "assignees": [
+                    _get_userinfo(assignee) for assignee in pull_request.assignees
+                ],
+                "repo": (
+                    pull_request.base.repo.full_name if pull_request.base else None
+                ),
+                "num_commits": str(pull_request.commits),
+                "num_files_changed": str(pull_request.changed_files),
+                "labels": [label.name for label in pull_request.labels],
+                "created_at": (
+                    pull_request.created_at.replace(tzinfo=timezone.utc)
+                    if pull_request.created_at
+                    else None
+                ),
+                "updated_at": (
+                    pull_request.updated_at.replace(tzinfo=timezone.utc)
+                    if pull_request.updated_at
+                    else None
+                ),
+                "closed_at": (
+                    pull_request.closed_at.replace(tzinfo=timezone.utc)
+                    if pull_request.closed_at
+                    else None
+                ),
+                "merged_at": (
+                    pull_request.merged_at.replace(tzinfo=timezone.utc)
+                    if pull_request.merged_at
+                    else None
+                ),
+                "merged_by": (
+                    _get_userinfo(pull_request.merged_by)
+                    if pull_request.merged_by
+                    else None
+                ),
+            }.items()
+            if v is not None
+        },
+    )
+
+
+def _fetch_issue_comments(issue: Issue) -> str:
+    comments = issue.get_comments()
+    return "\nComment: ".join(comment.body for comment in comments)
+
+
+def _convert_issue_to_document(
+    issue: Issue, repo_external_access: ExternalAccess | None
+) -> Document:
+    repo_name = issue.repository.full_name if issue.repository else ""
+    doc_metadata = DocMetadata(repo=repo_name)
+    file_content_byte = issue.body.encode('utf-8') if issue.body else b""
+    name = sanitize_filename(issue.title, "md")
+
+    return Document(
+        id=issue.html_url,
+        blob=file_content_byte,
+        source=DocumentSource.GITHUB,
+        extension=".md",
+        external_access=repo_external_access,
+        semantic_identifier=f"{issue.number}:{name}",
+        # updated_at is UTC time but is timezone unaware
+        doc_updated_at=issue.updated_at.replace(tzinfo=timezone.utc),
+        # this metadata is used in perm sync
+        doc_metadata=doc_metadata.model_dump(),
+        size_bytes=len(file_content_byte) if file_content_byte else 0,
+        primary_owners=[_get_userinfo(issue.user) if issue.user else None],
+        metadata={
+            k: [str(vi) for vi in v] if isinstance(v, list) else str(v)
+            for k, v in {
+                "object_type": "Issue",
+                "id": issue.number,
+                "state": issue.state,
+                "user": _get_userinfo(issue.user) if issue.user else None,
+                "assignees": [_get_userinfo(assignee) for assignee in issue.assignees],
+                "repo": issue.repository.full_name if issue.repository else None,
+                "labels": [label.name for label in issue.labels],
+                "created_at": (
+                    issue.created_at.replace(tzinfo=timezone.utc)
+                    if issue.created_at
+                    else None
+                ),
+                "updated_at": (
+                    issue.updated_at.replace(tzinfo=timezone.utc)
+                    if issue.updated_at
+                    else None
+                ),
+                "closed_at": (
+                    issue.closed_at.replace(tzinfo=timezone.utc)
+                    if issue.closed_at
+                    else None
+                ),
+                "closed_by": (
+                    _get_userinfo(issue.closed_by) if issue.closed_by else None
+                ),
+            }.items()
+            if v is not None
+        },
+    )
+
+
+class GithubConnectorStage(Enum):
+    START = "start"
+    PRS = "prs"
+    ISSUES = "issues"
+
+
+class GithubConnectorCheckpoint(ConnectorCheckpoint):
+    stage: GithubConnectorStage
+    curr_page: int
+
+    cached_repo_ids: list[int] | None = None
+    cached_repo: SerializedRepository | None = None
+
+    # Used for the fallback cursor-based pagination strategy
+    num_retrieved: int
+    cursor_url: str | None = None
+
+    def reset(self) -> None:
+        """
+        Resets curr_page, num_retrieved, and cursor_url to their initial values (0, 0, None)
+        """
+        self.curr_page = 0
+        self.num_retrieved = 0
+        self.cursor_url = None
+
+
+def make_cursor_url_callback(
+    checkpoint: GithubConnectorCheckpoint,
+) -> Callable[[str | None, int], None]:
+    def cursor_url_callback(cursor_url: str | None, num_objs: int) -> None:
+        # we want to maintain the old cursor url so code after retrieval
+        # can determine that we are using the fallback cursor-based pagination strategy
+        if cursor_url:
+            checkpoint.cursor_url = cursor_url
+        checkpoint.num_retrieved = num_objs
+
+    return cursor_url_callback
+
+
+class GithubConnector(CheckpointedConnectorWithPermSyncGH[GithubConnectorCheckpoint]):
+    def __init__(
+        self,
+        repo_owner: str,
+        repositories: str | None = None,
+        state_filter: str = "all",
+        include_prs: bool = True,
+        include_issues: bool = False,
+    ) -> None:
+        self.repo_owner = repo_owner
+        self.repositories = repositories
+        self.state_filter = state_filter
+        self.include_prs = include_prs
+        self.include_issues = include_issues
+        self.github_client: Github | None = None
+
+    def load_credentials(self, credentials: dict[str, Any]) -> dict[str, Any] | None:
+        # defaults to 30 items per page, can be set to as high as 100
+        token = credentials["github_access_token"]
+        auth = Auth.Token(token)
+
+        if GITHUB_CONNECTOR_BASE_URL:
+            self.github_client = Github(
+                auth=auth,
+                base_url=GITHUB_CONNECTOR_BASE_URL,
+                per_page=ITEMS_PER_PAGE,
+            )
+        else:
+            self.github_client = Github(
+                auth=auth,
+                per_page=ITEMS_PER_PAGE,
+            )
+
+        return None
+
+    def get_github_repo(
+        self, github_client: Github, attempt_num: int = 0
+    ) -> Repository.Repository:
+        if attempt_num > _MAX_NUM_RATE_LIMIT_RETRIES:
+            raise RuntimeError(
+                "Re-tried fetching repo too many times. Something is going wrong with fetching objects from Github"
+            )
+
+        try:
+            return github_client.get_repo(f"{self.repo_owner}/{self.repositories}")
+        except RateLimitExceededException:
+            sleep_after_rate_limit_exception(github_client)
+            return self.get_github_repo(github_client, attempt_num + 1)
+
+    def get_github_repos(
+        self, github_client: Github, attempt_num: int = 0
+    ) -> list[Repository.Repository]:
+        """Get specific repositories based on comma-separated repo_name string."""
+        if attempt_num > _MAX_NUM_RATE_LIMIT_RETRIES:
+            raise RuntimeError(
+                "Re-tried fetching repos too many times. Something is going wrong with fetching objects from Github"
+            )
+
+        try:
+            repos = []
+            # Split repo_name by comma and strip whitespace
+            repo_names = [
+                name.strip() for name in (cast(str, self.repositories)).split(",")
+            ]
+
+            for repo_name in repo_names:
+                if repo_name:  # Skip empty strings
+                    try:
+                        repo = github_client.get_repo(f"{self.repo_owner}/{repo_name}")
+                        repos.append(repo)
+                    except GithubException as e:
+                        logging.warning(
+                            f"Could not fetch repo {self.repo_owner}/{repo_name}: {e}"
+                        )
+
+            return repos
+        except RateLimitExceededException:
+            sleep_after_rate_limit_exception(github_client)
+            return self.get_github_repos(github_client, attempt_num + 1)
+
+    def get_all_repos(
+        self, github_client: Github, attempt_num: int = 0
+    ) -> list[Repository.Repository]:
+        if attempt_num > _MAX_NUM_RATE_LIMIT_RETRIES:
+            raise RuntimeError(
+                "Re-tried fetching repos too many times. Something is going wrong with fetching objects from Github"
+            )
+
+        try:
+            # Try to get organization first
+            try:
+                org = github_client.get_organization(self.repo_owner)
+                return list(org.get_repos())
+
+            except GithubException:
+                # If not an org, try as a user
+                user = github_client.get_user(self.repo_owner)
+                return list(user.get_repos())
+        except RateLimitExceededException:
+            sleep_after_rate_limit_exception(github_client)
+            return self.get_all_repos(github_client, attempt_num + 1)
+
+    def _pull_requests_func(
+        self, repo: Repository.Repository
+    ) -> Callable[[], PaginatedList[PullRequest]]:
+        return lambda: repo.get_pulls(
+            state=self.state_filter, sort="updated", direction="desc"
+        )
+
+    def _issues_func(
+        self, repo: Repository.Repository
+    ) -> Callable[[], PaginatedList[Issue]]:
+        return lambda: repo.get_issues(
+            state=self.state_filter, sort="updated", direction="desc"
+        )
+
+    def _fetch_from_github(
+        self,
+        checkpoint: GithubConnectorCheckpoint,
+        start: datetime | None = None,
+        end: datetime | None = None,
+        include_permissions: bool = False,
+    ) -> Generator[Document | ConnectorFailure, None, GithubConnectorCheckpoint]:
+        if self.github_client is None:
+            raise ConnectorMissingCredentialError("GitHub")
+
+        checkpoint = copy.deepcopy(checkpoint)
+
+        # First run of the connector, fetch all repos and store in checkpoint
+        if checkpoint.cached_repo_ids is None:
+            repos = []
+            if self.repositories:
+                if "," in self.repositories:
+                    # Multiple repositories specified
+                    repos = self.get_github_repos(self.github_client)
+                else:
+                    # Single repository (backward compatibility)
+                    repos = [self.get_github_repo(self.github_client)]
+            else:
+                # All repositories
+                repos = self.get_all_repos(self.github_client)
+            if not repos:
+                checkpoint.has_more = False
+                return checkpoint
+
+            curr_repo = repos.pop()
+            checkpoint.cached_repo_ids = [repo.id for repo in repos]
+            checkpoint.cached_repo = SerializedRepository(
+                id=curr_repo.id,
+                headers=curr_repo.raw_headers,
+                raw_data=curr_repo.raw_data,
+            )
+            checkpoint.stage = GithubConnectorStage.PRS
+            checkpoint.curr_page = 0
+            # save checkpoint with repo ids retrieved
+            return checkpoint
+
+        if checkpoint.cached_repo is None:
+            raise ValueError("No repo saved in checkpoint")
+
+        # Deserialize the repository from the checkpoint
+        repo = deserialize_repository(checkpoint.cached_repo, self.github_client)
+
+        cursor_url_callback = make_cursor_url_callback(checkpoint)
+        repo_external_access: ExternalAccess | None = None
+        if include_permissions:
+            repo_external_access = get_external_access_permission(
+                repo, self.github_client
+            )
+        if self.include_prs and checkpoint.stage == GithubConnectorStage.PRS:
+            logging.info(f"Fetching PRs for repo: {repo.name}")
+
+            pr_batch = _get_batch_rate_limited(
+                self._pull_requests_func(repo),
+                checkpoint.curr_page,
+                checkpoint.cursor_url,
+                checkpoint.num_retrieved,
+                cursor_url_callback,
+                self.github_client,
+            )
+            checkpoint.curr_page += 1  # NOTE: not used for cursor-based fallback
+            done_with_prs = False
+            num_prs = 0
+            pr = None
+            print("start: ", start)
+            for pr in pr_batch:
+                num_prs += 1
+                print("-"*40)
+                print("PR name", pr.title)
+                print("updated at", pr.updated_at)
+                print("-"*40)
+                print("\n")
+                # we iterate backwards in time, so at this point we stop processing prs
+                if (
+                    start is not None
+                    and pr.updated_at
+                    and pr.updated_at.replace(tzinfo=timezone.utc) <= start
+                ):
+                    done_with_prs = True
+                    break
+                # Skip PRs updated after the end date
+                if (
+                    end is not None
+                    and pr.updated_at
+                    and pr.updated_at.replace(tzinfo=timezone.utc) > end
+                ):
+                    continue
+                try:
+                    yield _convert_pr_to_document(
+                        cast(PullRequest, pr), repo_external_access
+                    )
+                except Exception as e:
+                    error_msg = f"Error converting PR to document: {e}"
+                    logging.exception(error_msg)
+                    yield ConnectorFailure(
+                        failed_document=DocumentFailure(
+                            document_id=str(pr.id), document_link=pr.html_url
+                        ),
+                        failure_message=error_msg,
+                        exception=e,
+                    )
+                    continue
+
+            # If we reach this point with a cursor url in the checkpoint, we were using
+            # the fallback cursor-based pagination strategy. That strategy tries to get all
+            # PRs, so having curosr_url set means we are done with prs. However, we need to
+            # return AFTER the checkpoint reset to avoid infinite loops.
+
+            # if we found any PRs on the page and there are more PRs to get, return the checkpoint.
+            # In offset mode, while indexing without time constraints, the pr batch
+            # will be empty when we're done.
+            used_cursor = checkpoint.cursor_url is not None
+            if num_prs > 0 and not done_with_prs and not used_cursor:
+                return checkpoint
+
+            # if we went past the start date during the loop or there are no more
+            # prs to get, we move on to issues
+            checkpoint.stage = GithubConnectorStage.ISSUES
+            checkpoint.reset()
+
+            if used_cursor:
+                # save the checkpoint after changing stage; next run will continue from issues
+                return checkpoint
+
+        checkpoint.stage = GithubConnectorStage.ISSUES
+
+        if self.include_issues and checkpoint.stage == GithubConnectorStage.ISSUES:
+            logging.info(f"Fetching issues for repo: {repo.name}")
+
+            issue_batch = list(
+                _get_batch_rate_limited(
+                    self._issues_func(repo),
+                    checkpoint.curr_page,
+                    checkpoint.cursor_url,
+                    checkpoint.num_retrieved,
+                    cursor_url_callback,
+                    self.github_client,
+                )
+            )
+            checkpoint.curr_page += 1
+            done_with_issues = False
+            num_issues = 0
+            for issue in issue_batch:
+                num_issues += 1
+                issue = cast(Issue, issue)
+                # we iterate backwards in time, so at this point we stop processing prs
+                if (
+                    start is not None
+                    and issue.updated_at.replace(tzinfo=timezone.utc) <= start
+                ):
+                    done_with_issues = True
+                    break
+                # Skip PRs updated after the end date
+                if (
+                    end is not None
+                    and issue.updated_at.replace(tzinfo=timezone.utc) > end
+                ):
+                    continue
+
+                if issue.pull_request is not None:
+                    # PRs are handled separately
+                    continue
+
+                try:
+                    yield _convert_issue_to_document(issue, repo_external_access)
+                except Exception as e:
+                    error_msg = f"Error converting issue to document: {e}"
+                    logging.exception(error_msg)
+                    yield ConnectorFailure(
+                        failed_document=DocumentFailure(
+                            document_id=str(issue.id),
+                            document_link=issue.html_url,
+                        ),
+                        failure_message=error_msg,
+                        exception=e,
+                    )
+                    continue
+
+            # if we found any issues on the page, and we're not done, return the checkpoint.
+            # don't return if we're using cursor-based pagination to avoid infinite loops
+            if num_issues > 0 and not done_with_issues and not checkpoint.cursor_url:
+                return checkpoint
+
+            # if we went past the start date during the loop or there are no more
+            # issues to get, we move on to the next repo
+            checkpoint.stage = GithubConnectorStage.PRS
+            checkpoint.reset()
+
+        checkpoint.has_more = len(checkpoint.cached_repo_ids) > 0
+        if checkpoint.cached_repo_ids:
+            next_id = checkpoint.cached_repo_ids.pop()
+            next_repo = self.github_client.get_repo(next_id)
+            checkpoint.cached_repo = SerializedRepository(
+                id=next_id,
+                headers=next_repo.raw_headers,
+                raw_data=next_repo.raw_data,
+            )
+            checkpoint.stage = GithubConnectorStage.PRS
+            checkpoint.reset()
+
+        if checkpoint.cached_repo_ids:
+            logging.info(
+                f"{len(checkpoint.cached_repo_ids)} repos remaining (IDs: {checkpoint.cached_repo_ids})"
+            )
+        else:
+            logging.info("No more repos remaining")
+
+        return checkpoint
+
+    def _load_from_checkpoint(
+        self,
+        start: SecondsSinceUnixEpoch,
+        end: SecondsSinceUnixEpoch,
+        checkpoint: GithubConnectorCheckpoint,
+        include_permissions: bool = False,
+    ) -> CheckpointOutput[GithubConnectorCheckpoint]:
+        start_datetime = datetime.fromtimestamp(start, tz=timezone.utc)
+        # add a day for timezone safety
+        end_datetime = datetime.fromtimestamp(end, tz=timezone.utc) + ONE_DAY
+
+        # Move start time back by 3 hours, since some Issues/PRs are getting dropped
+        # Could be due to delayed processing on GitHub side
+        # The non-updated issues since last poll will be shortcut-ed and not embedded
+        # adjusted_start_datetime = start_datetime - timedelta(hours=3)
+
+        adjusted_start_datetime = start_datetime
+
+        epoch = datetime.fromtimestamp(0, tz=timezone.utc)
+        if adjusted_start_datetime < epoch:
+            adjusted_start_datetime = epoch
+
+        return self._fetch_from_github(
+            checkpoint,
+            start=adjusted_start_datetime,
+            end=end_datetime,
+            include_permissions=include_permissions,
+        )
+
+    @override
+    def load_from_checkpoint(
+        self,
+        start: SecondsSinceUnixEpoch,
+        end: SecondsSinceUnixEpoch,
+        checkpoint: GithubConnectorCheckpoint,
+    ) -> CheckpointOutput[GithubConnectorCheckpoint]:
+        return self._load_from_checkpoint(
+            start, end, checkpoint, include_permissions=False
+        )
+
+    @override
+    def load_from_checkpoint_with_perm_sync(
+        self,
+        start: SecondsSinceUnixEpoch,
+        end: SecondsSinceUnixEpoch,
+        checkpoint: GithubConnectorCheckpoint,
+    ) -> CheckpointOutput[GithubConnectorCheckpoint]:
+        return self._load_from_checkpoint(
+            start, end, checkpoint, include_permissions=True
+        )
+
+    def validate_connector_settings(self) -> None:
+        if self.github_client is None:
+            raise ConnectorMissingCredentialError("GitHub credentials not loaded.")
+
+        if not self.repo_owner:
+            raise ConnectorValidationError(
+                "Invalid connector settings: 'repo_owner' must be provided."
+            )
+
+        try:
+            if self.repositories:
+                if "," in self.repositories:
+                    # Multiple repositories specified
+                    repo_names = [name.strip() for name in self.repositories.split(",")]
+                    if not repo_names:
+                        raise ConnectorValidationError(
+                            "Invalid connector settings: No valid repository names provided."
+                        )
+
+                    # Validate at least one repository exists and is accessible
+                    valid_repos = False
+                    validation_errors = []
+
+                    for repo_name in repo_names:
+                        if not repo_name:
+                            continue
+
+                        try:
+                            test_repo = self.github_client.get_repo(
+                                f"{self.repo_owner}/{repo_name}"
+                            )
+                            logging.info(
+                                f"Successfully accessed repository: {self.repo_owner}/{repo_name}"
+                            )
+                            test_repo.get_contents("")
+                            valid_repos = True
+                            # If at least one repo is valid, we can proceed
+                            break
+                        except GithubException as e:
+                            validation_errors.append(
+                                f"Repository '{repo_name}': {e.data.get('message', str(e))}"
+                            )
+
+                    if not valid_repos:
+                        error_msg = (
+                            "None of the specified repositories could be accessed: "
+                        )
+                        error_msg += ", ".join(validation_errors)
+                        raise ConnectorValidationError(error_msg)
+                else:
+                    # Single repository (backward compatibility)
+                    test_repo = self.github_client.get_repo(
+                        f"{self.repo_owner}/{self.repositories}"
+                    )
+                    test_repo.get_contents("")
+            else:
+                # Try to get organization first
+                try:
+                    org = self.github_client.get_organization(self.repo_owner)
+                    total_count = org.get_repos().totalCount
+                    if total_count == 0:
+                        raise ConnectorValidationError(
+                            f"Found no repos for organization: {self.repo_owner}. "
+                            "Does the credential have the right scopes?"
+                        )
+                except GithubException as e:
+                    # Check for missing SSO
+                    MISSING_SSO_ERROR_MESSAGE = "You must grant your Personal Access token access to this organization".lower()
+                    if MISSING_SSO_ERROR_MESSAGE in str(e).lower():
+                        SSO_GUIDE_LINK = (
+                            "https://docs.github.com/en/enterprise-cloud@latest/authentication/"
+                            "authenticating-with-saml-single-sign-on/"
+                            "authorizing-a-personal-access-token-for-use-with-saml-single-sign-on"
+                        )
+                        raise ConnectorValidationError(
+                            f"Your GitHub token is missing authorization to access the "
+                            f"`{self.repo_owner}` organization. Please follow the guide to "
+                            f"authorize your token: {SSO_GUIDE_LINK}"
+                        )
+                    # If not an org, try as a user
+                    user = self.github_client.get_user(self.repo_owner)
+
+                    # Check if we can access any repos
+                    total_count = user.get_repos().totalCount
+                    if total_count == 0:
+                        raise ConnectorValidationError(
+                            f"Found no repos for user: {self.repo_owner}. "
+                            "Does the credential have the right scopes?"
+                        )
+
+        except RateLimitExceededException:
+            raise UnexpectedValidationError(
+                "Validation failed due to GitHub rate-limits being exceeded. Please try again later."
+            )
+
+        except GithubException as e:
+            if e.status == 401:
+                raise CredentialExpiredError(
+                    "GitHub credential appears to be invalid or expired (HTTP 401)."
+                )
+            elif e.status == 403:
+                raise InsufficientPermissionsError(
+                    "Your GitHub token does not have sufficient permissions for this repository (HTTP 403)."
+                )
+            elif e.status == 404:
+                if self.repositories:
+                    if "," in self.repositories:
+                        raise ConnectorValidationError(
+                            f"None of the specified GitHub repositories could be found for owner: {self.repo_owner}"
+                        )
+                    else:
+                        raise ConnectorValidationError(
+                            f"GitHub repository not found with name: {self.repo_owner}/{self.repositories}"
+                        )
+                else:
+                    raise ConnectorValidationError(
+                        f"GitHub user or organization not found: {self.repo_owner}"
+                    )
+            else:
+                raise ConnectorValidationError(
+                    f"Unexpected GitHub error (status={e.status}): {e.data}"
+                )
+
+        except Exception as exc:
+            raise Exception(
+                f"Unexpected error during GitHub settings validation: {exc}"
+            )
+
+    def validate_checkpoint_json(
+        self, checkpoint_json: str
+    ) -> GithubConnectorCheckpoint:
+        return GithubConnectorCheckpoint.model_validate_json(checkpoint_json)
+
+    def build_dummy_checkpoint(self) -> GithubConnectorCheckpoint:
+        return GithubConnectorCheckpoint(
+            stage=GithubConnectorStage.PRS, curr_page=0, has_more=True, num_retrieved=0
+        )
+
+
+if __name__ == "__main__":
+    # Initialize the connector
+    connector = GithubConnector(
+        repo_owner="EvoAgentX",
+        repositories="EvoAgentX",
+        include_issues=True,
+        include_prs=False,
+    )
+    connector.load_credentials(
+        {"github_access_token": "<Your_GitHub_Access_Token>"}
+    )
+
+    if connector.github_client:
+        get_external_access_permission(
+            connector.get_github_repos(connector.github_client).pop(),
+            connector.github_client,
+        )
+
+    # Create a time range from epoch to now
+    end_time = datetime.now(timezone.utc)
+    start_time = datetime.fromtimestamp(0, tz=timezone.utc)
+    time_range = (start_time, end_time)
+
+    # Initialize the runner with a batch size of 10
+    runner: ConnectorRunner[GithubConnectorCheckpoint] = ConnectorRunner(
+        connector, batch_size=10, include_permissions=False, time_range=time_range
+    )
+
+    # Get initial checkpoint
+    checkpoint = connector.build_dummy_checkpoint()
+
+    # Run the connector
+    while checkpoint.has_more:
+        for doc_batch, failure, next_checkpoint in runner.run(checkpoint):
+            if doc_batch:
+                print(f"Retrieved batch of {len(doc_batch)} documents")
+                for doc in doc_batch:
+                    print(f"Document: {doc.semantic_identifier}")
+            if failure:
+                print(f"Failure: {failure.failure_message}")
+            if next_checkpoint:
+                checkpoint = next_checkpoint

common/data_source/github/models.py

@@ -0,0 +1,17 @@
+from typing import Any
+
+from github import Repository
+from github.Requester import Requester
+from pydantic import BaseModel
+
+
+class SerializedRepository(BaseModel):
+    # id is part of the raw_data as well, just pulled out for convenience
+    id: int
+    headers: dict[str, str | int]
+    raw_data: dict[str, Any]
+
+    def to_Repository(self, requester: Requester) -> Repository.Repository:
+        return Repository.Repository(
+            requester, self.headers, self.raw_data, completed=True
+        )

common/data_source/github/rate_limit_utils.py

@@ -0,0 +1,24 @@
+import time
+import logging
+from datetime import datetime
+from datetime import timedelta
+from datetime import timezone
+
+from github import Github
+
+
+def sleep_after_rate_limit_exception(github_client: Github) -> None:
+    """
+    Sleep until the GitHub rate limit resets.
+
+    Args:
+        github_client: The GitHub client that hit the rate limit
+    """
+    sleep_time = github_client.get_rate_limit().core.reset.replace(
+        tzinfo=timezone.utc
+    ) - datetime.now(tz=timezone.utc)
+    sleep_time += timedelta(minutes=1)  # add an extra minute just to be safe
+    logging.info(
+        "Ran into Github rate-limit. Sleeping %s seconds.", sleep_time.seconds
+    )
+    time.sleep(sleep_time.total_seconds())

common/data_source/github/utils.py

@@ -0,0 +1,44 @@
+import logging
+
+from github import Github
+from github.Repository import Repository
+
+from common.data_source.models import ExternalAccess
+
+from .models import SerializedRepository
+
+
+def get_external_access_permission(
+    repo: Repository, github_client: Github
+) -> ExternalAccess:
+    """
+    Get the external access permission for a repository.
+    This functionality requires Enterprise Edition.
+    """
+    # RAGFlow doesn't implement the Onyx EE external-permissions system.
+    # Default to private/unknown permissions.
+    return ExternalAccess.empty()
+
+
+def deserialize_repository(
+    cached_repo: SerializedRepository, github_client: Github
+) -> Repository:
+    """
+    Deserialize a SerializedRepository back into a Repository object.
+    """
+    # Try to access the requester - different PyGithub versions may use different attribute names
+    try:
+        # Try to get the requester using getattr to avoid linter errors
+        requester = getattr(github_client, "_requester", None)
+        if requester is None:
+            requester = getattr(github_client, "_Github__requester", None)
+        if requester is None:
+            # If we can't find the requester attribute, we need to fall back to recreating the repo
+            raise AttributeError("Could not find requester attribute")
+
+        return cached_repo.to_Repository(requester)
+    except Exception as e:
+        # If all else fails, re-fetch the repo directly
+        logging.warning("Failed to deserialize repository: %s. Attempting to re-fetch.", e)
+        repo_id = cached_repo.id
+        return github_client.get_repo(repo_id)

common/data_source/gitlab_connector.py

@@ -0,0 +1,340 @@
+import fnmatch
+import itertools
+from collections import deque
+from collections.abc import Iterable
+from collections.abc import Iterator
+from datetime import datetime
+from datetime import timezone
+from typing import Any
+from typing import TypeVar
+import gitlab
+from gitlab.v4.objects import Project
+
+from common.data_source.config import DocumentSource, INDEX_BATCH_SIZE
+from common.data_source.exceptions import ConnectorMissingCredentialError
+from common.data_source.exceptions import ConnectorValidationError
+from common.data_source.exceptions import CredentialExpiredError
+from common.data_source.exceptions import InsufficientPermissionsError
+from common.data_source.exceptions import UnexpectedValidationError
+from common.data_source.interfaces import GenerateDocumentsOutput
+from common.data_source.interfaces import LoadConnector
+from common.data_source.interfaces import PollConnector
+from common.data_source.interfaces import SecondsSinceUnixEpoch
+from common.data_source.models import BasicExpertInfo
+from common.data_source.models import Document
+from common.data_source.utils import get_file_ext
+
+T = TypeVar("T")
+
+
+
+# List of directories/Files to exclude
+exclude_patterns = [
+    "logs",
+    ".github/",
+    ".gitlab/",
+    ".pre-commit-config.yaml",
+]
+
+
+def _batch_gitlab_objects(git_objs: Iterable[T], batch_size: int) -> Iterator[list[T]]:
+    it = iter(git_objs)
+    while True:
+        batch = list(itertools.islice(it, batch_size))
+        if not batch:
+            break
+        yield batch
+
+
+def get_author(author: Any) -> BasicExpertInfo:
+    return BasicExpertInfo(
+        display_name=author.get("name"),
+    )
+
+
+def _convert_merge_request_to_document(mr: Any) -> Document:
+    mr_text = mr.description or ""
+    doc = Document(
+        id=mr.web_url,
+        blob=mr_text,
+        source=DocumentSource.GITLAB,
+        semantic_identifier=mr.title,
+        extension=".md",
+        # updated_at is UTC time but is timezone unaware, explicitly add UTC
+        # as there is logic in indexing to prevent wrong timestamped docs
+        # due to local time discrepancies with UTC
+        doc_updated_at=mr.updated_at.replace(tzinfo=timezone.utc),
+        size_bytes=len(mr_text.encode("utf-8")),
+        primary_owners=[get_author(mr.author)],
+        metadata={"state": mr.state, "type": "MergeRequest", "web_url": mr.web_url},
+    )
+    return doc
+
+
+def _convert_issue_to_document(issue: Any) -> Document:
+    issue_text = issue.description or ""
+    doc = Document(
+        id=issue.web_url,
+        blob=issue_text,
+        source=DocumentSource.GITLAB,
+        semantic_identifier=issue.title,
+        extension=".md",
+        # updated_at is UTC time but is timezone unaware, explicitly add UTC
+        # as there is logic in indexing to prevent wrong timestamped docs
+        # due to local time discrepancies with UTC
+        doc_updated_at=issue.updated_at.replace(tzinfo=timezone.utc),
+        size_bytes=len(issue_text.encode("utf-8")),
+        primary_owners=[get_author(issue.author)],
+        metadata={
+            "state": issue.state,
+            "type": issue.type if issue.type else "Issue",
+            "web_url": issue.web_url,
+        },
+    )
+    return doc
+
+
+def _convert_code_to_document(
+    project: Project, file: Any, url: str, projectName: str, projectOwner: str
+) -> Document:
+    
+    # Dynamically get the default branch from the project object
+    default_branch = project.default_branch
+
+    # Fetch the file content using the correct branch
+    file_content_obj = project.files.get(
+        file_path=file["path"], ref=default_branch  # Use the default branch
+    )
+    # BoxConnector uses raw bytes for blob. Keep the same here.
+    file_content_bytes = file_content_obj.decode()
+    file_url = f"{url}/{projectOwner}/{projectName}/-/blob/{default_branch}/{file['path']}"
+
+    # Try to use the last commit timestamp for incremental sync.
+    # Falls back to "now" if the commit lookup fails.
+    last_commit_at = None
+    try:
+        # Query commit history for this file on the default branch.
+        commits = project.commits.list(
+            ref_name=default_branch,
+            path=file["path"],
+            per_page=1,
+        )
+        if commits:
+            # committed_date is ISO string like "2024-01-01T00:00:00.000+00:00"
+            committed_date = commits[0].committed_date
+            if isinstance(committed_date, str):
+                last_commit_at = datetime.strptime(
+                    committed_date, "%Y-%m-%dT%H:%M:%S.%f%z"
+                ).astimezone(timezone.utc)
+            elif isinstance(committed_date, datetime):
+                last_commit_at = committed_date.astimezone(timezone.utc)
+    except Exception:
+        last_commit_at = None
+
+    # Create and return a Document object
+    doc = Document(
+        # Use a stable ID so reruns don't create duplicates.
+        id=file_url,
+        blob=file_content_bytes,
+        source=DocumentSource.GITLAB,
+        semantic_identifier=file.get("name"),
+        extension=get_file_ext(file.get("name")),
+        doc_updated_at=last_commit_at or datetime.now(tz=timezone.utc),
+        size_bytes=len(file_content_bytes) if file_content_bytes is not None else 0,
+        primary_owners=[],  # Add owners if needed
+        metadata={
+            "type": "CodeFile",
+            "path": file.get("path"),
+            "ref": default_branch,
+            "project": f"{projectOwner}/{projectName}",
+            "web_url": file_url,
+        },
+    )
+    return doc
+
+
+def _should_exclude(path: str) -> bool:
+    """Check if a path matches any of the exclude patterns."""
+    return any(fnmatch.fnmatch(path, pattern) for pattern in exclude_patterns)
+
+
+class GitlabConnector(LoadConnector, PollConnector):
+    def __init__(
+        self,
+        project_owner: str,
+        project_name: str,
+        batch_size: int = INDEX_BATCH_SIZE,
+        state_filter: str = "all",
+        include_mrs: bool = True,
+        include_issues: bool = True,
+        include_code_files: bool = False,
+    ) -> None:
+        self.project_owner = project_owner
+        self.project_name = project_name
+        self.batch_size = batch_size
+        self.state_filter = state_filter
+        self.include_mrs = include_mrs
+        self.include_issues = include_issues
+        self.include_code_files = include_code_files
+        self.gitlab_client: gitlab.Gitlab | None = None
+
+    def load_credentials(self, credentials: dict[str, Any]) -> dict[str, Any] | None:
+        self.gitlab_client = gitlab.Gitlab(
+            credentials["gitlab_url"], private_token=credentials["gitlab_access_token"]
+        )
+        return None
+
+    def validate_connector_settings(self) -> None:
+        if self.gitlab_client is None:
+            raise ConnectorMissingCredentialError("GitLab")
+
+        try:
+            self.gitlab_client.auth()
+            self.gitlab_client.projects.get(
+                f"{self.project_owner}/{self.project_name}",
+                lazy=True,
+            )
+
+        except gitlab.exceptions.GitlabAuthenticationError as e:
+            raise CredentialExpiredError(
+                "Invalid or expired GitLab credentials."
+            ) from e
+
+        except gitlab.exceptions.GitlabAuthorizationError as e:
+            raise InsufficientPermissionsError(
+                "Insufficient permissions to access GitLab resources."
+            ) from e
+
+        except gitlab.exceptions.GitlabGetError as e:
+            raise ConnectorValidationError(
+                "GitLab project not found or not accessible."
+            ) from e
+
+        except Exception as e:
+            raise UnexpectedValidationError(
+                f"Unexpected error while validating GitLab settings: {e}"
+            ) from e
+
+    def _fetch_from_gitlab(
+        self, start: datetime | None = None, end: datetime | None = None
+    ) -> GenerateDocumentsOutput:
+        if self.gitlab_client is None:
+            raise ConnectorMissingCredentialError("Gitlab")
+        project: Project = self.gitlab_client.projects.get(
+            f"{self.project_owner}/{self.project_name}"
+        )
+
+        start_utc = start.astimezone(timezone.utc) if start else None
+        end_utc = end.astimezone(timezone.utc) if end else None
+
+        # Fetch code files
+        if self.include_code_files:
+            # Fetching using BFS as project.report_tree with recursion causing slow load
+            queue = deque([""])  # Start with the root directory
+            while queue:
+                current_path = queue.popleft()
+                files = project.repository_tree(path=current_path, all=True)
+                for file_batch in _batch_gitlab_objects(files, self.batch_size):
+                    code_doc_batch: list[Document] = []
+                    for file in file_batch:
+                        if _should_exclude(file["path"]):
+                            continue
+
+                        if file["type"] == "blob":
+
+                            doc = _convert_code_to_document(
+                                project,
+                                file,
+                                self.gitlab_client.url,
+                                self.project_name,
+                                self.project_owner,
+                            )
+
+                            # Apply incremental window filtering for code files too.
+                            if start_utc is not None and doc.doc_updated_at <= start_utc:
+                                continue
+                            if end_utc is not None and doc.doc_updated_at > end_utc:
+                                continue
+
+                            code_doc_batch.append(doc)
+                        elif file["type"] == "tree":
+                            queue.append(file["path"])
+
+                    if code_doc_batch:
+                        yield code_doc_batch
+
+        if self.include_mrs:
+            merge_requests = project.mergerequests.list(
+                state=self.state_filter,
+                order_by="updated_at",
+                sort="desc",
+                iterator=True,
+            )
+
+            for mr_batch in _batch_gitlab_objects(merge_requests, self.batch_size):
+                mr_doc_batch: list[Document] = []
+                for mr in mr_batch:
+                    mr.updated_at = datetime.strptime(
+                        mr.updated_at, "%Y-%m-%dT%H:%M:%S.%f%z"
+                    )
+                    if start_utc is not None and mr.updated_at <= start_utc:
+                        yield mr_doc_batch
+                        return
+                    if end_utc is not None and mr.updated_at > end_utc:
+                        continue
+                    mr_doc_batch.append(_convert_merge_request_to_document(mr))
+                yield mr_doc_batch
+
+        if self.include_issues:
+            issues = project.issues.list(state=self.state_filter, iterator=True)
+
+            for issue_batch in _batch_gitlab_objects(issues, self.batch_size):
+                issue_doc_batch: list[Document] = []
+                for issue in issue_batch:
+                    issue.updated_at = datetime.strptime(
+                        issue.updated_at, "%Y-%m-%dT%H:%M:%S.%f%z"
+                    )
+                    # Avoid re-syncing the last-seen item.
+                    if start_utc is not None and issue.updated_at <= start_utc:
+                        yield issue_doc_batch
+                        return
+                    if end_utc is not None and issue.updated_at > end_utc:
+                        continue
+                    issue_doc_batch.append(_convert_issue_to_document(issue))
+                yield issue_doc_batch
+
+    def load_from_state(self) -> GenerateDocumentsOutput:
+        return self._fetch_from_gitlab()
+
+    def poll_source(
+        self, start: SecondsSinceUnixEpoch, end: SecondsSinceUnixEpoch
+    ) -> GenerateDocumentsOutput:
+        start_datetime = datetime.fromtimestamp(start, tz=timezone.utc)
+        end_datetime = datetime.fromtimestamp(end, tz=timezone.utc)
+        return self._fetch_from_gitlab(start_datetime, end_datetime)
+
+
+if __name__ == "__main__":
+    import os
+
+    connector = GitlabConnector(
+        # gitlab_url="https://gitlab.com/api/v4",
+        project_owner=os.environ["PROJECT_OWNER"],
+        project_name=os.environ["PROJECT_NAME"],
+        batch_size=INDEX_BATCH_SIZE,
+        state_filter="all",
+        include_mrs=True,
+        include_issues=True,
+        include_code_files=True,
+    )
+
+    connector.load_credentials(
+        {
+            "gitlab_access_token": os.environ["GITLAB_ACCESS_TOKEN"],
+            "gitlab_url": os.environ["GITLAB_URL"],
+        }
+    )
+    document_batches = connector.load_from_state()
+    for f in document_batches:
+        print("Batch:", f)
+    print("Finished loading from state.")

common/data_source/google_util/util.py

@@ -191,7 +191,7 @@ def get_credentials_from_env(email: str, oauth: bool = False, source="drive") ->
         DB_CREDENTIALS_AUTHENTICATION_METHOD: "uploaded",
     }
 
-def sanitize_filename(name: str) -> str:
+def sanitize_filename(name: str, extension: str = "txt") -> str:
     """
     Soft sanitize for MinIO/S3:
     - Replace only prohibited characters with a space.
@@ -199,7 +199,7 @@ def sanitize_filename(name: str) -> str:
     - Collapse multiple spaces.
     """
     if name is None:
-        return "file.txt"
+        return f"file.{extension}"
 
     name = str(name).strip()
 
@@ -222,9 +222,8 @@ def sanitize_filename(name: str) -> str:
         base, ext = os.path.splitext(name)
         name = base[:180].rstrip() + ext
 
-    # Ensure there is an extension (your original logic)
     if not os.path.splitext(name)[1]:
-        name += ".txt"
+        name += f".{extension}"
 
     return name
 

common/data_source/imap_connector.py

@@ -0,0 +1,724 @@
+import copy
+import email
+from email.header import decode_header
+import imaplib
+import logging
+import os
+import re
+from datetime import datetime, timedelta
+from datetime import timezone
+from email.message import Message
+from email.utils import collapse_rfc2231_value, parseaddr
+from enum import Enum
+from typing import Any
+from typing import cast
+
+import bs4
+from pydantic import BaseModel
+
+from common.data_source.config import IMAP_CONNECTOR_SIZE_THRESHOLD, DocumentSource
+from common.data_source.interfaces import CheckpointOutput, CheckpointedConnectorWithPermSync, CredentialsConnector, CredentialsProviderInterface
+from common.data_source.models import BasicExpertInfo, ConnectorCheckpoint, Document, ExternalAccess, SecondsSinceUnixEpoch
+
+_DEFAULT_IMAP_PORT_NUMBER = int(os.environ.get("IMAP_PORT", 993))
+_IMAP_OKAY_STATUS = "OK"
+_PAGE_SIZE = 100
+_USERNAME_KEY = "imap_username"
+_PASSWORD_KEY = "imap_password"
+
+class Header(str, Enum):
+    SUBJECT_HEADER = "subject"
+    FROM_HEADER = "from"
+    TO_HEADER = "to"
+    CC_HEADER = "cc"
+    DELIVERED_TO_HEADER = (
+        "Delivered-To"  # Used in mailing lists instead of the "to" header.
+    )
+    DATE_HEADER = "date"
+    MESSAGE_ID_HEADER = "Message-ID"
+
+
+class EmailHeaders(BaseModel):
+    """
+    Model for email headers extracted from IMAP messages.
+    """
+
+    id: str
+    subject: str
+    sender: str
+    recipients: str | None
+    cc: str | None
+    date: datetime
+
+    @classmethod
+    def from_email_msg(cls, email_msg: Message) -> "EmailHeaders":
+        def _decode(header: str, default: str | None = None) -> str | None:
+            value = email_msg.get(header, default)
+            if not value:
+                return None
+
+            decoded_fragments = decode_header(value)
+            decoded_strings: list[str] = []
+
+            for decoded_value, encoding in decoded_fragments:
+                if isinstance(decoded_value, bytes):
+                    try:
+                        decoded_strings.append(
+                            decoded_value.decode(encoding or "utf-8", errors="replace")
+                        )
+                    except LookupError:
+                        decoded_strings.append(
+                            decoded_value.decode("utf-8", errors="replace")
+                        )
+                elif isinstance(decoded_value, str):
+                    decoded_strings.append(decoded_value)
+                else:
+                    decoded_strings.append(str(decoded_value))
+
+            return "".join(decoded_strings)
+
+        def _parse_date(date_str: str | None) -> datetime | None:
+            if not date_str:
+                return None
+            try:
+                return email.utils.parsedate_to_datetime(date_str)
+            except (TypeError, ValueError):
+                return None
+
+        message_id = _decode(header=Header.MESSAGE_ID_HEADER)
+        if not message_id:
+            message_id = f"<generated-{uuid.uuid4()}@imap.local>"
+        # It's possible for the subject line to not exist or be an empty string.
+        subject = _decode(header=Header.SUBJECT_HEADER) or "Unknown Subject"
+        from_ = _decode(header=Header.FROM_HEADER)
+        to = _decode(header=Header.TO_HEADER)
+        if not to:
+            to = _decode(header=Header.DELIVERED_TO_HEADER)
+        cc = _decode(header=Header.CC_HEADER)
+        date_str = _decode(header=Header.DATE_HEADER)
+        date = _parse_date(date_str=date_str)
+
+        if not date:
+            date = datetime.now(tz=timezone.utc)
+
+        # If any of the above are `None`, model validation will fail.
+        # Therefore, no guards (i.e.: `if <header> is None: raise RuntimeError(..)`) were written.
+        return cls.model_validate(
+            {
+                "id": message_id,
+                "subject": subject,
+                "sender": from_,
+                "recipients": to,
+                "cc": cc,
+                "date": date,
+            }
+        )
+
+class CurrentMailbox(BaseModel):
+    mailbox: str
+    todo_email_ids: list[str]
+
+
+# An email has a list of mailboxes.
+# Each mailbox has a list of email-ids inside of it.
+#
+# Usage:
+# To use this checkpointer, first fetch all the mailboxes.
+# Then, pop a mailbox and fetch all of its email-ids.
+# Then, pop each email-id and fetch its content (and parse it, etc..).
+# When you have popped all email-ids for this mailbox, pop the next mailbox and repeat the above process until you're done.
+#
+# For initial checkpointing, set both fields to `None`.
+class ImapCheckpoint(ConnectorCheckpoint):
+    todo_mailboxes: list[str] | None = None
+    current_mailbox: CurrentMailbox | None = None
+
+
+class LoginState(str, Enum):
+    LoggedIn = "logged_in"
+    LoggedOut = "logged_out"
+
+
+class ImapConnector(
+    CredentialsConnector,
+    CheckpointedConnectorWithPermSync,
+):
+    def __init__(
+        self,
+        host: str,
+        port: int = _DEFAULT_IMAP_PORT_NUMBER,
+        mailboxes: list[str] | None = None,
+    ) -> None:
+        self._host = host
+        self._port = port
+        self._mailboxes = mailboxes
+        self._credentials: dict[str, Any] | None = None
+
+    @property
+    def credentials(self) -> dict[str, Any]:
+        if not self._credentials:
+            raise RuntimeError(
+                "Credentials have not been initialized; call `set_credentials_provider` first"
+            )
+        return self._credentials
+
+    def _get_mail_client(self) -> imaplib.IMAP4_SSL:
+        """
+        Returns a new `imaplib.IMAP4_SSL` instance.
+
+        The `imaplib.IMAP4_SSL` object is supposed to be an "ephemeral" object; it's not something that you can login,
+        logout, then log back into again. I.e., the following will fail:
+
+        ```py
+        mail_client.login(..)
+        mail_client.logout();
+        mail_client.login(..)
+        ```
+
+        Therefore, you need a fresh, new instance in order to operate with IMAP. This function gives one to you.
+
+        # Notes
+        This function will throw an error if the credentials have not yet been set.
+        """
+
+        def get_or_raise(name: str) -> str:
+            value = self.credentials.get(name)
+            if not value:
+                raise RuntimeError(f"Credential item {name=} was not found")
+            if not isinstance(value, str):
+                raise RuntimeError(
+                    f"Credential item {name=} must be of type str, instead received {type(name)=}"
+                )
+            return value
+
+        username = get_or_raise(_USERNAME_KEY)
+        password = get_or_raise(_PASSWORD_KEY)
+
+        mail_client = imaplib.IMAP4_SSL(host=self._host, port=self._port)
+        status, _data = mail_client.login(user=username, password=password)
+
+        if status != _IMAP_OKAY_STATUS:
+            raise RuntimeError(f"Failed to log into imap server; {status=}")
+
+        return mail_client
+
+    def _load_from_checkpoint(
+        self,
+        start: SecondsSinceUnixEpoch,
+        end: SecondsSinceUnixEpoch,
+        checkpoint: ImapCheckpoint,
+        include_perm_sync: bool,
+    ) -> CheckpointOutput[ImapCheckpoint]:
+        checkpoint = cast(ImapCheckpoint, copy.deepcopy(checkpoint))
+        checkpoint.has_more = True
+
+        mail_client = self._get_mail_client()
+
+        if checkpoint.todo_mailboxes is None:
+            # This is the dummy checkpoint.
+            # Fill it with mailboxes first.
+            if self._mailboxes:
+                checkpoint.todo_mailboxes = _sanitize_mailbox_names(self._mailboxes)
+            else:
+                fetched_mailboxes = _fetch_all_mailboxes_for_email_account(
+                    mail_client=mail_client
+                )
+                if not fetched_mailboxes:
+                    raise RuntimeError(
+                        "Failed to find any mailboxes for this email account"
+                    )
+                checkpoint.todo_mailboxes = _sanitize_mailbox_names(fetched_mailboxes)
+
+            return checkpoint
+
+        if (
+            not checkpoint.current_mailbox
+            or not checkpoint.current_mailbox.todo_email_ids
+        ):
+            if not checkpoint.todo_mailboxes:
+                checkpoint.has_more = False
+                return checkpoint
+
+            mailbox = checkpoint.todo_mailboxes.pop()
+            email_ids = _fetch_email_ids_in_mailbox(
+                mail_client=mail_client,
+                mailbox=mailbox,
+                start=start,
+                end=end,
+            )
+            checkpoint.current_mailbox = CurrentMailbox(
+                mailbox=mailbox,
+                todo_email_ids=email_ids,
+            )
+
+        _select_mailbox(
+            mail_client=mail_client, mailbox=checkpoint.current_mailbox.mailbox
+        )
+        current_todos = cast(
+            list, copy.deepcopy(checkpoint.current_mailbox.todo_email_ids[:_PAGE_SIZE])
+        )
+        checkpoint.current_mailbox.todo_email_ids = (
+            checkpoint.current_mailbox.todo_email_ids[_PAGE_SIZE:]
+        )
+
+        for email_id in current_todos:
+            email_msg = _fetch_email(mail_client=mail_client, email_id=email_id)
+            if not email_msg:
+                logging.warning(f"Failed to fetch message {email_id=}; skipping")
+                continue
+
+            email_headers = EmailHeaders.from_email_msg(email_msg=email_msg)
+            msg_dt = email_headers.date
+            if msg_dt.tzinfo is None:
+                msg_dt = msg_dt.replace(tzinfo=timezone.utc)
+            else:
+                msg_dt = msg_dt.astimezone(timezone.utc)
+
+            start_dt = datetime.fromtimestamp(start, tz=timezone.utc)
+            end_dt = datetime.fromtimestamp(end, tz=timezone.utc)
+
+            if not (start_dt < msg_dt <= end_dt):
+                continue
+
+            email_doc = _convert_email_headers_and_body_into_document(
+                email_msg=email_msg,
+                email_headers=email_headers,
+                include_perm_sync=include_perm_sync,
+            )
+            yield email_doc
+            attachments = extract_attachments(email_msg)
+            for att in attachments:
+                yield attachment_to_document(email_doc, att, email_headers)
+
+        return checkpoint
+
+    # impls for BaseConnector
+
+    def load_credentials(self, credentials: dict[str, Any]) -> dict[str, Any] | None:
+        self._credentials = credentials
+        return None
+
+    def validate_connector_settings(self) -> None:
+        self._get_mail_client()
+
+    # impls for CredentialsConnector
+
+    def set_credentials_provider(
+        self, credentials_provider: CredentialsProviderInterface
+    ) -> None:
+        self._credentials = credentials_provider.get_credentials()
+
+    # impls for CheckpointedConnector
+
+    def load_from_checkpoint(
+        self,
+        start: SecondsSinceUnixEpoch,
+        end: SecondsSinceUnixEpoch,
+        checkpoint: ImapCheckpoint,
+    ) -> CheckpointOutput[ImapCheckpoint]:
+        return self._load_from_checkpoint(
+            start=start, end=end, checkpoint=checkpoint, include_perm_sync=False
+        )
+
+    def build_dummy_checkpoint(self) -> ImapCheckpoint:
+        return ImapCheckpoint(has_more=True)
+
+    def validate_checkpoint_json(self, checkpoint_json: str) -> ImapCheckpoint:
+        return ImapCheckpoint.model_validate_json(json_data=checkpoint_json)
+
+    # impls for CheckpointedConnectorWithPermSync
+
+    def load_from_checkpoint_with_perm_sync(
+        self,
+        start: SecondsSinceUnixEpoch,
+        end: SecondsSinceUnixEpoch,
+        checkpoint: ImapCheckpoint,
+    ) -> CheckpointOutput[ImapCheckpoint]:
+        return self._load_from_checkpoint(
+            start=start, end=end, checkpoint=checkpoint, include_perm_sync=True
+        )
+
+
+def _fetch_all_mailboxes_for_email_account(mail_client: imaplib.IMAP4_SSL) -> list[str]:
+    status, mailboxes_data = mail_client.list('""', "*")
+    if status != _IMAP_OKAY_STATUS:
+        raise RuntimeError(f"Failed to fetch mailboxes; {status=}")
+
+    mailboxes = []
+
+    for mailboxes_raw in mailboxes_data:
+        if isinstance(mailboxes_raw, bytes):
+            mailboxes_str = mailboxes_raw.decode()
+        elif isinstance(mailboxes_raw, str):
+            mailboxes_str = mailboxes_raw
+        else:
+            logging.warning(
+                f"Expected the mailbox data to be of type str, instead got {type(mailboxes_raw)=} {mailboxes_raw}; skipping"
+            )
+            continue
+
+        # The mailbox LIST response output can be found here:
+        # https://www.rfc-editor.org/rfc/rfc3501.html#section-7.2.2
+        #
+        # The general format is:
+        # `(<name-attributes>) <hierarchy-delimiter> <mailbox-name>`
+        #
+        # The below regex matches on that pattern; from there, we select the 3rd match (index 2), which is the mailbox-name.
+        match = re.match(r'\([^)]*\)\s+"([^"]+)"\s+"?(.+?)"?$', mailboxes_str)
+        if not match:
+            logging.warning(
+                f"Invalid mailbox-data formatting structure: {mailboxes_str=}; skipping"
+            )
+            continue
+
+        mailbox = match.group(2)
+        mailboxes.append(mailbox)
+    if not mailboxes:
+        logging.warning(
+            "No mailboxes parsed from LIST response; falling back to INBOX"
+        )
+        return ["INBOX"]
+
+    return mailboxes
+
+
+def _select_mailbox(mail_client: imaplib.IMAP4_SSL, mailbox: str) -> bool:
+    try:
+        status, _ = mail_client.select(mailbox=mailbox, readonly=True)
+        if status != _IMAP_OKAY_STATUS:
+            return False
+        return True
+    except Exception:
+        return False
+
+
+def _fetch_email_ids_in_mailbox(
+    mail_client: imaplib.IMAP4_SSL,
+    mailbox: str,
+    start: SecondsSinceUnixEpoch,
+    end: SecondsSinceUnixEpoch,
+) -> list[str]:
+    if not _select_mailbox(mail_client, mailbox):
+        logging.warning(f"Skip mailbox: {mailbox}")
+        return []
+
+    start_dt = datetime.fromtimestamp(start, tz=timezone.utc)
+    end_dt = datetime.fromtimestamp(end, tz=timezone.utc) + timedelta(days=1)
+
+    start_str = start_dt.strftime("%d-%b-%Y")
+    end_str = end_dt.strftime("%d-%b-%Y")
+    search_criteria = f'(SINCE "{start_str}" BEFORE "{end_str}")'
+
+    status, email_ids_byte_array = mail_client.search(None, search_criteria)
+
+    if status != _IMAP_OKAY_STATUS or not email_ids_byte_array:
+        raise RuntimeError(f"Failed to fetch email ids; {status=}")
+
+    email_ids: bytes = email_ids_byte_array[0]
+
+    return [email_id.decode() for email_id in email_ids.split()]
+
+
+def _fetch_email(mail_client: imaplib.IMAP4_SSL, email_id: str) -> Message | None:
+    status, msg_data = mail_client.fetch(message_set=email_id, message_parts="(RFC822)")
+    if status != _IMAP_OKAY_STATUS or not msg_data:
+        return None
+
+    data = msg_data[0]
+    if not isinstance(data, tuple):
+        raise RuntimeError(
+            f"Message data should be a tuple; instead got a {type(data)=} {data=}"
+        )
+
+    _, raw_email = data
+    return email.message_from_bytes(raw_email)
+
+
+def _convert_email_headers_and_body_into_document(
+    email_msg: Message,
+    email_headers: EmailHeaders,
+    include_perm_sync: bool,
+) -> Document:
+    sender_name, sender_addr = _parse_singular_addr(raw_header=email_headers.sender)
+    to_addrs = (
+        _parse_addrs(email_headers.recipients)
+        if email_headers.recipients
+        else []
+    )
+    cc_addrs = (
+        _parse_addrs(email_headers.cc)
+        if email_headers.cc
+        else []
+    )
+    all_participants = to_addrs + cc_addrs
+
+    expert_info_map = {
+        recipient_addr: BasicExpertInfo(
+            display_name=recipient_name, email=recipient_addr
+        )
+        for recipient_name, recipient_addr in all_participants
+    }
+    if sender_addr not in expert_info_map:
+        expert_info_map[sender_addr] = BasicExpertInfo(
+            display_name=sender_name, email=sender_addr
+        )
+
+    email_body = _parse_email_body(email_msg=email_msg, email_headers=email_headers)
+    primary_owners = list(expert_info_map.values())
+    external_access = (
+        ExternalAccess(
+            external_user_emails=set(expert_info_map.keys()),
+            external_user_group_ids=set(),
+            is_public=False,
+        )
+        if include_perm_sync
+        else None
+    )
+    return Document(
+        id=email_headers.id,
+        title=email_headers.subject,
+        blob=email_body,
+        size_bytes=len(email_body),
+        semantic_identifier=email_headers.subject,
+        metadata={},
+        extension='.txt',
+        doc_updated_at=email_headers.date,
+        source=DocumentSource.IMAP,
+        primary_owners=primary_owners,
+        external_access=external_access,
+    )
+
+def extract_attachments(email_msg: Message, max_bytes: int = IMAP_CONNECTOR_SIZE_THRESHOLD):
+    attachments = []
+
+    if not email_msg.is_multipart():
+        return attachments
+
+    for part in email_msg.walk():
+        if part.get_content_maintype() == "multipart":
+            continue
+
+        disposition = (part.get("Content-Disposition") or "").lower()
+        filename = part.get_filename()
+
+        if not (
+            disposition.startswith("attachment")
+            or (disposition.startswith("inline") and filename)
+        ):
+            continue
+
+        payload = part.get_payload(decode=True)
+        if not payload:
+            continue
+
+        if len(payload) > max_bytes:
+            continue
+
+        attachments.append({
+            "filename": filename or "attachment.bin",
+            "content_type": part.get_content_type(),
+            "content_bytes": payload,
+            "size_bytes": len(payload),
+        })
+
+    return attachments
+
+def decode_mime_filename(raw: str | None) -> str | None:
+    if not raw:
+        return None
+
+    try:
+        raw = collapse_rfc2231_value(raw)
+    except Exception:
+        pass
+
+    parts = decode_header(raw)
+    decoded = []
+
+    for value, encoding in parts:
+        if isinstance(value, bytes):
+            decoded.append(value.decode(encoding or "utf-8", errors="replace"))
+        else:
+            decoded.append(value)
+
+    return "".join(decoded)
+
+def attachment_to_document(
+    parent_doc: Document,
+    att: dict,
+    email_headers: EmailHeaders,
+):
+    raw_filename = att["filename"]
+    filename = decode_mime_filename(raw_filename) or "attachment.bin"
+    ext = "." + filename.split(".")[-1] if "." in filename else ""
+
+    return Document(
+        id=f"{parent_doc.id}#att:{filename}",
+        source=DocumentSource.IMAP,
+        semantic_identifier=filename,
+        extension=ext,
+        blob=att["content_bytes"],
+        size_bytes=att["size_bytes"],
+        doc_updated_at=email_headers.date,
+        primary_owners=parent_doc.primary_owners,
+        metadata={
+            "parent_email_id": parent_doc.id,
+            "parent_subject": email_headers.subject,
+            "attachment_filename": filename,
+            "attachment_content_type": att["content_type"],
+        },
+    )
+
+def _parse_email_body(
+    email_msg: Message,
+    email_headers: EmailHeaders,
+) -> str:
+    body = None
+    for part in email_msg.walk():
+        if part.is_multipart():
+            # Multipart parts are *containers* for other parts, not the actual content itself.
+            # Therefore, we skip until we find the individual parts instead.
+            continue
+
+        charset = part.get_content_charset() or "utf-8"
+
+        try:
+            raw_payload = part.get_payload(decode=True)
+            if not isinstance(raw_payload, bytes):
+                logging.warning(
+                    "Payload section from email was expected to be an array of bytes, instead got "
+                    f"{type(raw_payload)=}, {raw_payload=}"
+                )
+                continue
+            body = raw_payload.decode(charset)
+            break
+        except (UnicodeDecodeError, LookupError) as e:
+            logging.warning(f"Could not decode part with charset {charset}. Error: {e}")
+            continue
+
+    if not body:
+        logging.warning(
+            f"Email with {email_headers.id=} has an empty body; returning an empty string"
+        )
+        return ""
+
+    soup = bs4.BeautifulSoup(markup=body, features="html.parser")
+
+    return " ".join(str_section for str_section in soup.stripped_strings)
+
+
+def _sanitize_mailbox_names(mailboxes: list[str]) -> list[str]:
+    """
+    Mailboxes with special characters in them must be enclosed by double-quotes, as per the IMAP protocol.
+    Just to be safe, we wrap *all* mailboxes with double-quotes.
+    """
+    return [f'"{mailbox}"' for mailbox in mailboxes if mailbox]
+
+
+def _parse_addrs(raw_header: str) -> list[tuple[str, str]]:
+    addrs = raw_header.split(",")
+    name_addr_pairs = [parseaddr(addr=addr) for addr in addrs if addr]
+    return [(name, addr) for name, addr in name_addr_pairs if addr]
+
+
+def _parse_singular_addr(raw_header: str) -> tuple[str, str]:
+    addrs = _parse_addrs(raw_header=raw_header)
+    if not addrs:
+        return ("Unknown", "unknown@example.com")
+    elif len(addrs) >= 2:
+        raise RuntimeError(
+            f"Expected a singular address, but instead got multiple; {raw_header=} {addrs=}"
+        )
+
+    return addrs[0]
+
+
+if __name__ == "__main__":
+    import time
+    import uuid
+    from types import TracebackType
+    from common.data_source.utils import load_all_docs_from_checkpoint_connector
+
+
+    class OnyxStaticCredentialsProvider(
+        CredentialsProviderInterface["OnyxStaticCredentialsProvider"]
+    ):
+        """Implementation (a very simple one!) to handle static credentials."""
+
+        def __init__(
+            self,
+            tenant_id: str | None,
+            connector_name: str,
+            credential_json: dict[str, Any],
+        ):
+            self._tenant_id = tenant_id
+            self._connector_name = connector_name
+            self._credential_json = credential_json
+
+            self._provider_key = str(uuid.uuid4())
+
+        def __enter__(self) -> "OnyxStaticCredentialsProvider":
+            return self
+
+        def __exit__(
+            self,
+            exc_type: type[BaseException] | None,
+            exc_value: BaseException | None,
+            traceback: TracebackType | None,
+        ) -> None:
+            pass
+
+        def get_tenant_id(self) -> str | None:
+            return self._tenant_id
+
+        def get_provider_key(self) -> str:
+            return self._provider_key
+
+        def get_credentials(self) -> dict[str, Any]:
+            return self._credential_json
+
+        def set_credentials(self, credential_json: dict[str, Any]) -> None:
+            self._credential_json = credential_json
+
+        def is_dynamic(self) -> bool:
+            return False
+    # from tests.daily.connectors.utils import load_all_docs_from_checkpoint_connector
+    # from onyx.connectors.credentials_provider import OnyxStaticCredentialsProvider
+
+    host = os.environ.get("IMAP_HOST")
+    mailboxes_str = os.environ.get("IMAP_MAILBOXES","INBOX")
+    username = os.environ.get("IMAP_USERNAME")
+    password = os.environ.get("IMAP_PASSWORD")
+
+    mailboxes = (
+        [mailbox.strip() for mailbox in mailboxes_str.split(",")]
+        if mailboxes_str
+        else []
+    )
+
+    if not host:
+        raise RuntimeError("`IMAP_HOST` must be set")
+
+    imap_connector = ImapConnector(
+        host=host,
+        mailboxes=mailboxes,
+    )
+
+    imap_connector.set_credentials_provider(
+        OnyxStaticCredentialsProvider(
+            tenant_id=None,
+            connector_name=DocumentSource.IMAP,
+            credential_json={
+                _USERNAME_KEY: username,
+                _PASSWORD_KEY: password,
+            },
+        )
+    )
+    END = time.time()
+    START = END - 1 * 24 * 60 * 60
+    for doc in load_all_docs_from_checkpoint_connector(
+        connector=imap_connector,
+        start=START,
+        end=END,
+    ):
+        print(doc.id,doc.extension)

common/data_source/interfaces.py

@@ -5,7 +5,7 @@ from abc import ABC, abstractmethod
 from enum import IntFlag, auto
 from types import TracebackType
 from typing import Any, Dict, Generator, TypeVar, Generic, Callable, TypeAlias
-
+from collections.abc import Iterator
 from anthropic import BaseModel
 
 from common.data_source.models import (
@@ -16,6 +16,7 @@ from common.data_source.models import (
     SecondsSinceUnixEpoch, GenerateSlimDocumentOutput
 )
 
+GenerateDocumentsOutput = Iterator[list[Document]]
 
 class LoadConnector(ABC):
     """Load connector interface"""
@@ -236,16 +237,13 @@ class BaseConnector(abc.ABC, Generic[CT]):
 
     def validate_perm_sync(self) -> None:
         """
-        Don't override this; add a function to perm_sync_valid.py in the ee package
-        to do permission sync validation
+        Permission-sync validation hook.
+
+        RAGFlow doesn't ship the Onyx EE permission-sync validation package.
+        Connectors that support permission sync should override
+        `validate_connector_settings()` as needed.
         """
-        """
-        validate_connector_settings_fn = fetch_ee_implementation_or_noop(
-            "onyx.connectors.perm_sync_valid",
-            "validate_perm_sync",
-            noop_return_value=None,
-        )
-        validate_connector_settings_fn(self)"""
+        return None
 
     def set_allow_images(self, value: bool) -> None:
         """Implement if the underlying connector wants to skip/allow image downloading
@@ -344,6 +342,17 @@ class CheckpointOutputWrapper(Generic[CT]):
         yield None, None, self.next_checkpoint
 
 
+class CheckpointedConnectorWithPermSyncGH(CheckpointedConnector[CT]):
+    @abc.abstractmethod
+    def load_from_checkpoint_with_perm_sync(
+        self,
+        start: SecondsSinceUnixEpoch,
+        end: SecondsSinceUnixEpoch,
+        checkpoint: CT,
+    ) -> CheckpointOutput[CT]:
+        raise NotImplementedError
+
+
 # Slim connectors retrieve just the ids of documents
 class SlimConnector(BaseConnector):
     @abc.abstractmethod

common/data_source/models.py

@@ -94,8 +94,10 @@ class Document(BaseModel):
     blob: bytes
     doc_updated_at: datetime
     size_bytes: int
-    primary_owners: list
+    externale_access: Optional[ExternalAccess] = None
+    primary_owners: Optional[list] = None
     metadata: Optional[dict[str, Any]] = None
+    doc_metadata: Optional[dict[str, Any]] = None
 
 
 class BasicExpertInfo(BaseModel):

common/doc_store/doc_store_base.py

@@ -13,7 +13,6 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 #
-
 from abc import ABC, abstractmethod
 from dataclasses import dataclass
 import numpy as np
@@ -22,7 +21,6 @@ DEFAULT_MATCH_VECTOR_TOPN = 10
 DEFAULT_MATCH_SPARSE_TOPN = 10
 VEC = list | np.ndarray
 
-
 @dataclass
 class SparseVector:
     indices: list[int]
@@ -55,14 +53,13 @@ class SparseVector:
     def __repr__(self):
         return str(self)
 
-
-class MatchTextExpr(ABC):
+class MatchTextExpr:
     def __init__(
         self,
         fields: list[str],
         matching_text: str,
         topn: int,
-        extra_options: dict = dict(),
+        extra_options: dict | None = None,
     ):
         self.fields = fields
         self.matching_text = matching_text
@@ -70,7 +67,7 @@ class MatchTextExpr(ABC):
         self.extra_options = extra_options
 
 
-class MatchDenseExpr(ABC):
+class MatchDenseExpr:
     def __init__(
         self,
         vector_column_name: str,
@@ -78,7 +75,7 @@ class MatchDenseExpr(ABC):
         embedding_data_type: str,
         distance_type: str,
         topn: int = DEFAULT_MATCH_VECTOR_TOPN,
-        extra_options: dict = dict(),
+        extra_options: dict | None = None,
     ):
         self.vector_column_name = vector_column_name
         self.embedding_data = embedding_data
@@ -88,7 +85,7 @@ class MatchDenseExpr(ABC):
         self.extra_options = extra_options
 
 
-class MatchSparseExpr(ABC):
+class MatchSparseExpr:
     def __init__(
         self,
         vector_column_name: str,
@@ -104,7 +101,7 @@ class MatchSparseExpr(ABC):
         self.opt_params = opt_params
 
 
-class MatchTensorExpr(ABC):
+class MatchTensorExpr:
     def __init__(
         self,
         column_name: str,
@@ -120,7 +117,7 @@ class MatchTensorExpr(ABC):
         self.extra_option = extra_option
 
 
-class FusionExpr(ABC):
+class FusionExpr:
     def __init__(self, method: str, topn: int, fusion_params: dict | None = None):
         self.method = method
         self.topn = topn
@@ -129,7 +126,8 @@ class FusionExpr(ABC):
 
 MatchExpr = MatchTextExpr | MatchDenseExpr | MatchSparseExpr | MatchTensorExpr | FusionExpr
 
-class OrderByExpr(ABC):
+
+class OrderByExpr:
     def __init__(self):
         self.fields = list()
     def asc(self, field: str):
@@ -141,13 +139,14 @@ class OrderByExpr(ABC):
     def fields(self):
         return self.fields
 
+
 class DocStoreConnection(ABC):
     """
     Database operations
     """
 
     @abstractmethod
-    def dbType(self) -> str:
+    def db_type(self) -> str:
         """
         Return the type of the database.
         """
@@ -165,21 +164,21 @@ class DocStoreConnection(ABC):
     """
 
     @abstractmethod
-    def createIdx(self, indexName: str, knowledgebaseId: str, vectorSize: int):
+    def create_idx(self, index_name: str, dataset_id: str, vector_size: int):
         """
         Create an index with given name
         """
         raise NotImplementedError("Not implemented")
 
     @abstractmethod
-    def deleteIdx(self, indexName: str, knowledgebaseId: str):
+    def delete_idx(self, index_name: str, dataset_id: str):
         """
         Delete an index with given name
         """
         raise NotImplementedError("Not implemented")
 
     @abstractmethod
-    def indexExist(self, indexName: str, knowledgebaseId: str) -> bool:
+    def index_exist(self, index_name: str, dataset_id: str) -> bool:
         """
         Check if an index with given name exists
         """
@@ -191,16 +190,16 @@ class DocStoreConnection(ABC):
 
     @abstractmethod
     def search(
-        self, selectFields: list[str],
-            highlightFields: list[str],
+        self, select_fields: list[str],
+            highlight_fields: list[str],
             condition: dict,
-            matchExprs: list[MatchExpr],
-            orderBy: OrderByExpr,
+            match_expressions: list[MatchExpr],
+            order_by: OrderByExpr,
             offset: int,
             limit: int,
-            indexNames: str|list[str],
-            knowledgebaseIds: list[str],
-            aggFields: list[str] = [],
+            index_names: str|list[str],
+            dataset_ids: list[str],
+            agg_fields: list[str] | None = None,
             rank_feature: dict | None = None
     ):
         """
@@ -209,28 +208,28 @@ class DocStoreConnection(ABC):
         raise NotImplementedError("Not implemented")
 
     @abstractmethod
-    def get(self, chunkId: str, indexName: str, knowledgebaseIds: list[str]) -> dict | None:
+    def get(self, data_id: str, index_name: str, dataset_ids: list[str]) -> dict | None:
         """
         Get single chunk with given id
         """
         raise NotImplementedError("Not implemented")
 
     @abstractmethod
-    def insert(self, rows: list[dict], indexName: str, knowledgebaseId: str = None) -> list[str]:
+    def insert(self, rows: list[dict], index_name: str, dataset_id: str = None) -> list[str]:
         """
         Update or insert a bulk of rows
         """
         raise NotImplementedError("Not implemented")
 
     @abstractmethod
-    def update(self, condition: dict, newValue: dict, indexName: str, knowledgebaseId: str) -> bool:
+    def update(self, condition: dict, new_value: dict, index_name: str, dataset_id: str) -> bool:
         """
         Update rows with given conjunctive equivalent filtering condition
         """
         raise NotImplementedError("Not implemented")
 
     @abstractmethod
-    def delete(self, condition: dict, indexName: str, knowledgebaseId: str) -> int:
+    def delete(self, condition: dict, index_name: str, dataset_id: str) -> int:
         """
         Delete rows with given conjunctive equivalent filtering condition
         """
@@ -245,7 +244,7 @@ class DocStoreConnection(ABC):
         raise NotImplementedError("Not implemented")
 
     @abstractmethod
-    def get_chunk_ids(self, res):
+    def get_doc_ids(self, res):
         raise NotImplementedError("Not implemented")
 
     @abstractmethod
@@ -253,18 +252,18 @@ class DocStoreConnection(ABC):
         raise NotImplementedError("Not implemented")
 
     @abstractmethod
-    def get_highlight(self, res, keywords: list[str], fieldnm: str):
+    def get_highlight(self, res, keywords: list[str], field_name: str):
         raise NotImplementedError("Not implemented")
 
     @abstractmethod
-    def get_aggregation(self, res, fieldnm: str):
+    def get_aggregation(self, res, field_name: str):
         raise NotImplementedError("Not implemented")
 
     """
     SQL
     """
     @abstractmethod
-    def sql(sql: str, fetch_size: int, format: str):
+    def sql(self, sql: str, fetch_size: int, format: str):
         """
         Run the sql generated by text-to-sql
         """

common/doc_store/es_conn_base.py

@@ -0,0 +1,307 @@
+#
+#  Copyright 2025 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 re
+import json
+import time
+import os
+from abc import abstractmethod
+
+from elasticsearch import NotFoundError
+from elasticsearch_dsl import Index
+from elastic_transport import ConnectionTimeout
+from common.file_utils import get_project_base_directory
+from common.misc_utils import convert_bytes
+from common.doc_store.doc_store_base import DocStoreConnection, OrderByExpr, MatchExpr
+from rag.nlp import is_english, rag_tokenizer
+from common import settings
+
+ATTEMPT_TIME = 2
+
+
+class ESConnectionBase(DocStoreConnection):
+    def __init__(self, mapping_file_name: str="mapping.json", logger_name: str='ragflow.es_conn'):
+        from common.doc_store.es_conn_pool import ES_CONN
+
+        self.logger = logging.getLogger(logger_name)
+
+        self.info = {}
+        self.logger.info(f"Use Elasticsearch {settings.ES['hosts']} as the doc engine.")
+        self.es = ES_CONN.get_conn()
+        fp_mapping = os.path.join(get_project_base_directory(), "conf", mapping_file_name)
+        if not os.path.exists(fp_mapping):
+            msg = f"Elasticsearch mapping file not found at {fp_mapping}"
+            self.logger.error(msg)
+            raise Exception(msg)
+        self.mapping = json.load(open(fp_mapping, "r"))
+        self.logger.info(f"Elasticsearch {settings.ES['hosts']} is healthy.")
+
+    def _connect(self):
+        from common.doc_store.es_conn_pool import ES_CONN
+
+        if self.es.ping():
+            return True
+        self.es = ES_CONN.refresh_conn()
+        return True
+
+    """
+    Database operations
+    """
+
+    def db_type(self) -> str:
+        return "elasticsearch"
+
+    def health(self) -> dict:
+        health_dict = dict(self.es.cluster.health())
+        health_dict["type"] = "elasticsearch"
+        return health_dict
+
+    def get_cluster_stats(self):
+        """
+        curl -XGET "http://{es_host}/_cluster/stats" -H "kbn-xsrf: reporting" to view raw stats.
+        """
+        raw_stats = self.es.cluster.stats()
+        self.logger.debug(f"ESConnection.get_cluster_stats: {raw_stats}")
+        try:
+            res = {
+                'cluster_name': raw_stats['cluster_name'],
+                'status': raw_stats['status']
+            }
+            indices_status = raw_stats['indices']
+            res.update({
+                'indices': indices_status['count'],
+                'indices_shards': indices_status['shards']['total']
+            })
+            doc_info = indices_status['docs']
+            res.update({
+                'docs': doc_info['count'],
+                'docs_deleted': doc_info['deleted']
+            })
+            store_info = indices_status['store']
+            res.update({
+                'store_size': convert_bytes(store_info['size_in_bytes']),
+                'total_dataset_size': convert_bytes(store_info['total_data_set_size_in_bytes'])
+            })
+            mappings_info = indices_status['mappings']
+            res.update({
+                'mappings_fields': mappings_info['total_field_count'],
+                'mappings_deduplicated_fields': mappings_info['total_deduplicated_field_count'],
+                'mappings_deduplicated_size': convert_bytes(mappings_info['total_deduplicated_mapping_size_in_bytes'])
+            })
+            node_info = raw_stats['nodes']
+            res.update({
+                'nodes': node_info['count']['total'],
+                'nodes_version': node_info['versions'],
+                'os_mem': convert_bytes(node_info['os']['mem']['total_in_bytes']),
+                'os_mem_used': convert_bytes(node_info['os']['mem']['used_in_bytes']),
+                'os_mem_used_percent': node_info['os']['mem']['used_percent'],
+                'jvm_versions': node_info['jvm']['versions'][0]['vm_version'],
+                'jvm_heap_used': convert_bytes(node_info['jvm']['mem']['heap_used_in_bytes']),
+                'jvm_heap_max': convert_bytes(node_info['jvm']['mem']['heap_max_in_bytes'])
+            })
+            return res
+
+        except Exception as e:
+            self.logger.exception(f"ESConnection.get_cluster_stats: {e}")
+            return None
+
+    """
+    Table operations
+    """
+
+    def create_idx(self, index_name: str, dataset_id: str, vector_size: int):
+        if self.index_exist(index_name, dataset_id):
+            return True
+        try:
+            from elasticsearch.client import IndicesClient
+            return IndicesClient(self.es).create(index=index_name,
+                                                 settings=self.mapping["settings"],
+                                                 mappings=self.mapping["mappings"])
+        except Exception:
+            self.logger.exception("ESConnection.createIndex error %s" % index_name)
+
+    def delete_idx(self, index_name: str, dataset_id: str):
+        if len(dataset_id) > 0:
+            # The index need to be alive after any kb deletion since all kb under this tenant are in one index.
+            return
+        try:
+            self.es.indices.delete(index=index_name, allow_no_indices=True)
+        except NotFoundError:
+            pass
+        except Exception:
+            self.logger.exception("ESConnection.deleteIdx error %s" % index_name)
+
+    def index_exist(self, index_name: str, dataset_id: str = None) -> bool:
+        s = Index(index_name, self.es)
+        for i in range(ATTEMPT_TIME):
+            try:
+                return s.exists()
+            except ConnectionTimeout:
+                self.logger.exception("ES request timeout")
+                time.sleep(3)
+                self._connect()
+                continue
+            except Exception as e:
+                self.logger.exception(e)
+                break
+        return False
+
+    """
+    CRUD operations
+    """
+
+    def get(self, doc_id: str, index_name: str, dataset_ids: list[str]) -> dict | None:
+        for i in range(ATTEMPT_TIME):
+            try:
+                res = self.es.get(index=index_name,
+                                  id=doc_id, source=True, )
+                if str(res.get("timed_out", "")).lower() == "true":
+                    raise Exception("Es Timeout.")
+                doc = res["_source"]
+                doc["id"] = doc_id
+                return doc
+            except NotFoundError:
+                return None
+            except Exception as e:
+                self.logger.exception(f"ESConnection.get({doc_id}) got exception")
+                raise e
+        self.logger.error(f"ESConnection.get timeout for {ATTEMPT_TIME} times!")
+        raise Exception("ESConnection.get timeout.")
+
+    @abstractmethod
+    def search(
+            self, select_fields: list[str],
+            highlight_fields: list[str],
+            condition: dict,
+            match_expressions: list[MatchExpr],
+            order_by: OrderByExpr,
+            offset: int,
+            limit: int,
+            index_names: str | list[str],
+            dataset_ids: list[str],
+            agg_fields: list[str] | None = None,
+            rank_feature: dict | None = None
+    ):
+        raise NotImplementedError("Not implemented")
+
+    @abstractmethod
+    def insert(self, documents: list[dict], index_name: str, dataset_id: str = None) -> list[str]:
+        raise NotImplementedError("Not implemented")
+
+    @abstractmethod
+    def update(self, condition: dict, new_value: dict, index_name: str, dataset_id: str) -> bool:
+        raise NotImplementedError("Not implemented")
+
+    @abstractmethod
+    def delete(self, condition: dict, index_name: str, dataset_id: str) -> int:
+        raise NotImplementedError("Not implemented")
+
+    """
+    Helper functions for search result
+    """
+
+    def get_total(self, res):
+        if isinstance(res["hits"]["total"], type({})):
+            return res["hits"]["total"]["value"]
+        return res["hits"]["total"]
+
+    def get_doc_ids(self, res):
+        return [d["_id"] for d in res["hits"]["hits"]]
+
+    def _get_source(self, res):
+        rr = []
+        for d in res["hits"]["hits"]:
+            d["_source"]["id"] = d["_id"]
+            d["_source"]["_score"] = d["_score"]
+            rr.append(d["_source"])
+        return rr
+
+    @abstractmethod
+    def get_fields(self, res, fields: list[str]) -> dict[str, dict]:
+        raise NotImplementedError("Not implemented")
+
+    def get_highlight(self, res, keywords: list[str], field_name: str):
+        ans = {}
+        for d in res["hits"]["hits"]:
+            highlights = d.get("highlight")
+            if not highlights:
+                continue
+            txt = "...".join([a for a in list(highlights.items())[0][1]])
+            if not is_english(txt.split()):
+                ans[d["_id"]] = txt
+                continue
+
+            txt = d["_source"][field_name]
+            txt = re.sub(r"[\r\n]", " ", txt, flags=re.IGNORECASE | re.MULTILINE)
+            txt_list = []
+            for t in re.split(r"[.?!;\n]", txt):
+                for w in keywords:
+                    t = re.sub(r"(^|[ .?/'\"\(\)!,:;-])(%s)([ .?/'\"\(\)!,:;-])" % re.escape(w), r"\1<em>\2</em>\3", t,
+                               flags=re.IGNORECASE | re.MULTILINE)
+                if not re.search(r"<em>[^<>]+</em>", t, flags=re.IGNORECASE | re.MULTILINE):
+                    continue
+                txt_list.append(t)
+            ans[d["_id"]] = "...".join(txt_list) if txt_list else "...".join([a for a in list(highlights.items())[0][1]])
+
+        return ans
+
+    def get_aggregation(self, res, field_name: str):
+        agg_field = "aggs_" + field_name
+        if "aggregations" not in res or agg_field not in res["aggregations"]:
+            return list()
+        buckets = res["aggregations"][agg_field]["buckets"]
+        return [(b["key"], b["doc_count"]) for b in buckets]
+
+    """
+    SQL
+    """
+
+    def sql(self, sql: str, fetch_size: int, format: str):
+        self.logger.debug(f"ESConnection.sql get sql: {sql}")
+        sql = re.sub(r"[ `]+", " ", sql)
+        sql = sql.replace("%", "")
+        replaces = []
+        for r in re.finditer(r" ([a-z_]+_l?tks)( like | ?= ?)'([^']+)'", sql):
+            fld, v = r.group(1), r.group(3)
+            match = " MATCH({}, '{}', 'operator=OR;minimum_should_match=30%') ".format(
+                fld, rag_tokenizer.fine_grained_tokenize(rag_tokenizer.tokenize(v)))
+            replaces.append(
+                ("{}{}'{}'".format(
+                    r.group(1),
+                    r.group(2),
+                    r.group(3)),
+                 match))
+
+        for p, r in replaces:
+            sql = sql.replace(p, r, 1)
+        self.logger.debug(f"ESConnection.sql to es: {sql}")
+
+        for i in range(ATTEMPT_TIME):
+            try:
+                res = self.es.sql.query(body={"query": sql, "fetch_size": fetch_size}, format=format,
+                                        request_timeout="2s")
+                return res
+            except ConnectionTimeout:
+                self.logger.exception("ES request timeout")
+                time.sleep(3)
+                self._connect()
+                continue
+            except Exception as e:
+                self.logger.exception(f"ESConnection.sql got exception. SQL:\n{sql}")
+                raise Exception(f"SQL error: {e}\n\nSQL: {sql}")
+        self.logger.error(f"ESConnection.sql timeout for {ATTEMPT_TIME} times!")
+        return None

common/doc_store/es_conn_pool.py

@@ -0,0 +1,84 @@
+#
+#  Copyright 2025 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 time
+from elasticsearch import Elasticsearch
+
+from common import settings
+from common.decorator import singleton
+
+ATTEMPT_TIME = 2
+
+
+@singleton
+class ElasticSearchConnectionPool:
+
+    def __init__(self):
+        if hasattr(settings, "ES"):
+            self.ES_CONFIG = settings.ES
+        else:
+            self.ES_CONFIG = settings.get_base_config("es", {})
+
+        for _ in range(ATTEMPT_TIME):
+            try:
+                if self._connect():
+                    break
+            except Exception as e:
+                logging.warning(f"{str(e)}. Waiting Elasticsearch {self.ES_CONFIG['hosts']} to be healthy.")
+                time.sleep(5)
+
+        if not hasattr(self, "es_conn") or not self.es_conn or not self.es_conn.ping():
+            msg = f"Elasticsearch {self.ES_CONFIG['hosts']} is unhealthy in 10s."
+            logging.error(msg)
+            raise Exception(msg)
+        v = self.info.get("version", {"number": "8.11.3"})
+        v = v["number"].split(".")[0]
+        if int(v) < 8:
+            msg = f"Elasticsearch version must be greater than or equal to 8, current version: {v}"
+            logging.error(msg)
+            raise Exception(msg)
+
+    def _connect(self):
+        self.es_conn = Elasticsearch(
+            self.ES_CONFIG["hosts"].split(","),
+            basic_auth=(self.ES_CONFIG["username"], self.ES_CONFIG[
+                "password"]) if "username" in self.ES_CONFIG and "password" in self.ES_CONFIG else None,
+            verify_certs= self.ES_CONFIG.get("verify_certs", False),
+            timeout=600 )
+        if self.es_conn:
+            self.info = self.es_conn.info()
+            return True
+        return False
+
+    def get_conn(self):
+        return self.es_conn
+
+    def refresh_conn(self):
+        if self.es_conn.ping():
+            return self.es_conn
+        else:
+            # close current if exist
+            if self.es_conn:
+                self.es_conn.close()
+            self._connect()
+            return self.es_conn
+
+    def __del__(self):
+        if hasattr(self, "es_conn") and self.es_conn:
+            self.es_conn.close()
+
+
+ES_CONN = ElasticSearchConnectionPool()

common/doc_store/infinity_conn_base.py

@@ -0,0 +1,453 @@
+#
+#  Copyright 2025 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 re
+import json
+import time
+from abc import abstractmethod
+
+import infinity
+from infinity.common import ConflictType
+from infinity.index import IndexInfo, IndexType
+from infinity.errors import ErrorCode
+import pandas as pd
+from common.file_utils import get_project_base_directory
+from rag.nlp import is_english
+from common import settings
+from common.doc_store.doc_store_base import DocStoreConnection, MatchExpr, OrderByExpr
+
+
+class InfinityConnectionBase(DocStoreConnection):
+    def __init__(self, mapping_file_name: str="infinity_mapping.json", logger_name: str="ragflow.infinity_conn"):
+        from common.doc_store.infinity_conn_pool import INFINITY_CONN
+
+        self.dbName = settings.INFINITY.get("db_name", "default_db")
+        self.mapping_file_name = mapping_file_name
+        self.logger = logging.getLogger(logger_name)
+        infinity_uri = settings.INFINITY["uri"]
+        if ":" in infinity_uri:
+            host, port = infinity_uri.split(":")
+            infinity_uri = infinity.common.NetworkAddress(host, int(port))
+        self.connPool = None
+        self.logger.info(f"Use Infinity {infinity_uri} as the doc engine.")
+        conn_pool = INFINITY_CONN.get_conn_pool()
+        for _ in range(24):
+            try:
+                inf_conn = conn_pool.get_conn()
+                res = inf_conn.show_current_node()
+                if res.error_code == ErrorCode.OK and res.server_status in ["started", "alive"]:
+                    self._migrate_db(inf_conn)
+                    self.connPool = conn_pool
+                    conn_pool.release_conn(inf_conn)
+                    break
+                conn_pool.release_conn(inf_conn)
+                self.logger.warning(f"Infinity status: {res.server_status}. Waiting Infinity {infinity_uri} to be healthy.")
+                time.sleep(5)
+            except Exception as e:
+                conn_pool = INFINITY_CONN.refresh_conn_pool()
+                self.logger.warning(f"{str(e)}. Waiting Infinity {infinity_uri} to be healthy.")
+                time.sleep(5)
+        if self.connPool is None:
+            msg = f"Infinity {infinity_uri} is unhealthy in 120s."
+            self.logger.error(msg)
+            raise Exception(msg)
+        self.logger.info(f"Infinity {infinity_uri} is healthy.")
+
+    def _migrate_db(self, inf_conn):
+        inf_db = inf_conn.create_database(self.dbName, ConflictType.Ignore)
+        fp_mapping = os.path.join(get_project_base_directory(), "conf", self.mapping_file_name)
+        if not os.path.exists(fp_mapping):
+            raise Exception(f"Mapping file not found at {fp_mapping}")
+        schema = json.load(open(fp_mapping))
+        table_names = inf_db.list_tables().table_names
+        for table_name in table_names:
+            inf_table = inf_db.get_table(table_name)
+            index_names = inf_table.list_indexes().index_names
+            if "q_vec_idx" not in index_names:
+                # Skip tables not created by me
+                continue
+            column_names = inf_table.show_columns()["name"]
+            column_names = set(column_names)
+            for field_name, field_info in schema.items():
+                if field_name in column_names:
+                    continue
+                res = inf_table.add_columns({field_name: field_info})
+                assert res.error_code == infinity.ErrorCode.OK
+                self.logger.info(f"INFINITY added following column to table {table_name}: {field_name} {field_info}")
+                if field_info["type"] != "varchar" or "analyzer" not in field_info:
+                    continue
+                analyzers = field_info["analyzer"]
+                if isinstance(analyzers, str):
+                    analyzers = [analyzers]
+                for analyzer in analyzers:
+                    inf_table.create_index(
+                        f"ft_{re.sub(r'[^a-zA-Z0-9]', '_', field_name)}_{re.sub(r'[^a-zA-Z0-9]', '_', analyzer)}",
+                        IndexInfo(field_name, IndexType.FullText, {"ANALYZER": analyzer}),
+                        ConflictType.Ignore,
+                    )
+
+    """
+    Dataframe and fields convert
+    """
+
+    @staticmethod
+    @abstractmethod
+    def field_keyword(field_name: str):
+        # judge keyword or not, such as "*_kwd" tag-like columns.
+        raise NotImplementedError("Not implemented")
+
+    @abstractmethod
+    def convert_select_fields(self, output_fields: list[str]) -> list[str]:
+        # rm _kwd, _tks, _sm_tks, _with_weight suffix in field name.
+        raise NotImplementedError("Not implemented")
+
+    @staticmethod
+    @abstractmethod
+    def convert_matching_field(field_weight_str: str) -> str:
+        # convert matching field to
+        raise NotImplementedError("Not implemented")
+
+    @staticmethod
+    def list2str(lst: str | list, sep: str = " ") -> str:
+        if isinstance(lst, str):
+            return lst
+        return sep.join(lst)
+
+    def equivalent_condition_to_str(self, condition: dict, table_instance=None) -> str | None:
+        assert "_id" not in condition
+        columns = {}
+        if table_instance:
+            for n, ty, de, _ in table_instance.show_columns().rows():
+                columns[n] = (ty, de)
+
+        def exists(cln):
+            nonlocal columns
+            assert cln in columns, f"'{cln}' should be in '{columns}'."
+            ty, de = columns[cln]
+            if ty.lower().find("cha"):
+                if not de:
+                    de = ""
+                return f" {cln}!='{de}' "
+            return f"{cln}!={de}"
+
+        cond = list()
+        for k, v in condition.items():
+            if not isinstance(k, str) or not v:
+                continue
+            if self.field_keyword(k):
+                if isinstance(v, list):
+                    inCond = list()
+                    for item in v:
+                        if isinstance(item, str):
+                            item = item.replace("'", "''")
+                        inCond.append(f"filter_fulltext('{self.convert_matching_field(k)}', '{item}')")
+                    if inCond:
+                        strInCond = " or ".join(inCond)
+                        strInCond = f"({strInCond})"
+                        cond.append(strInCond)
+                else:
+                    cond.append(f"filter_fulltext('{self.convert_matching_field(k)}', '{v}')")
+            elif isinstance(v, list):
+                inCond = list()
+                for item in v:
+                    if isinstance(item, str):
+                        item = item.replace("'", "''")
+                        inCond.append(f"'{item}'")
+                    else:
+                        inCond.append(str(item))
+                if inCond:
+                    strInCond = ", ".join(inCond)
+                    strInCond = f"{k} IN ({strInCond})"
+                    cond.append(strInCond)
+            elif k == "must_not":
+                if isinstance(v, dict):
+                    for kk, vv in v.items():
+                        if kk == "exists":
+                            cond.append("NOT (%s)" % exists(vv))
+            elif isinstance(v, str):
+                cond.append(f"{k}='{v}'")
+            elif k == "exists":
+                cond.append(exists(v))
+            else:
+                cond.append(f"{k}={str(v)}")
+        return " AND ".join(cond) if cond else "1=1"
+
+    @staticmethod
+    def concat_dataframes(df_list: list[pd.DataFrame], select_fields: list[str]) -> pd.DataFrame:
+        df_list2 = [df for df in df_list if not df.empty]
+        if df_list2:
+            return pd.concat(df_list2, axis=0).reset_index(drop=True)
+
+        schema = []
+        for field_name in select_fields:
+            if field_name == "score()":  # Workaround: fix schema is changed to score()
+                schema.append("SCORE")
+            elif field_name == "similarity()":  # Workaround: fix schema is changed to similarity()
+                schema.append("SIMILARITY")
+            else:
+                schema.append(field_name)
+        return pd.DataFrame(columns=schema)
+
+    """
+    Database operations
+    """
+
+    def db_type(self) -> str:
+        return "infinity"
+
+    def health(self) -> dict:
+        """
+        Return the health status of the database.
+        """
+        inf_conn = self.connPool.get_conn()
+        res = inf_conn.show_current_node()
+        self.connPool.release_conn(inf_conn)
+        res2 = {
+            "type": "infinity",
+            "status": "green" if res.error_code == 0 and res.server_status in ["started", "alive"] else "red",
+            "error": res.error_msg,
+        }
+        return res2
+
+    """
+    Table operations
+    """
+
+    def create_idx(self, index_name: str, dataset_id: str, vector_size: int):
+        table_name = f"{index_name}_{dataset_id}"
+        inf_conn = self.connPool.get_conn()
+        inf_db = inf_conn.create_database(self.dbName, ConflictType.Ignore)
+
+        fp_mapping = os.path.join(get_project_base_directory(), "conf", self.mapping_file_name)
+        if not os.path.exists(fp_mapping):
+            raise Exception(f"Mapping file not found at {fp_mapping}")
+        schema = json.load(open(fp_mapping))
+        vector_name = f"q_{vector_size}_vec"
+        schema[vector_name] = {"type": f"vector,{vector_size},float"}
+        inf_table = inf_db.create_table(
+            table_name,
+            schema,
+            ConflictType.Ignore,
+        )
+        inf_table.create_index(
+            "q_vec_idx",
+            IndexInfo(
+                vector_name,
+                IndexType.Hnsw,
+                {
+                    "M": "16",
+                    "ef_construction": "50",
+                    "metric": "cosine",
+                    "encode": "lvq",
+                },
+            ),
+            ConflictType.Ignore,
+        )
+        for field_name, field_info in schema.items():
+            if field_info["type"] != "varchar" or "analyzer" not in field_info:
+                continue
+            analyzers = field_info["analyzer"]
+            if isinstance(analyzers, str):
+                analyzers = [analyzers]
+            for analyzer in analyzers:
+                inf_table.create_index(
+                    f"ft_{re.sub(r'[^a-zA-Z0-9]', '_', field_name)}_{re.sub(r'[^a-zA-Z0-9]', '_', analyzer)}",
+                    IndexInfo(field_name, IndexType.FullText, {"ANALYZER": analyzer}),
+                    ConflictType.Ignore,
+                )
+        self.connPool.release_conn(inf_conn)
+        self.logger.info(f"INFINITY created table {table_name}, vector size {vector_size}")
+        return True
+
+    def delete_idx(self, index_name: str, dataset_id: str):
+        table_name = f"{index_name}_{dataset_id}"
+        inf_conn = self.connPool.get_conn()
+        db_instance = inf_conn.get_database(self.dbName)
+        db_instance.drop_table(table_name, ConflictType.Ignore)
+        self.connPool.release_conn(inf_conn)
+        self.logger.info(f"INFINITY dropped table {table_name}")
+
+    def index_exist(self, index_name: str, dataset_id: str) -> bool:
+        table_name = f"{index_name}_{dataset_id}"
+        try:
+            inf_conn = self.connPool.get_conn()
+            db_instance = inf_conn.get_database(self.dbName)
+            _ = db_instance.get_table(table_name)
+            self.connPool.release_conn(inf_conn)
+            return True
+        except Exception as e:
+            self.logger.warning(f"INFINITY indexExist {str(e)}")
+        return False
+
+    """
+    CRUD operations
+    """
+
+    @abstractmethod
+    def search(
+        self,
+        select_fields: list[str],
+        highlight_fields: list[str],
+        condition: dict,
+        match_expressions: list[MatchExpr],
+        order_by: OrderByExpr,
+        offset: int,
+        limit: int,
+        index_names: str | list[str],
+        dataset_ids: list[str],
+        agg_fields: list[str] | None = None,
+        rank_feature: dict | None = None,
+    ) -> tuple[pd.DataFrame, int]:
+        raise NotImplementedError("Not implemented")
+
+    @abstractmethod
+    def get(self, doc_id: str, index_name: str, knowledgebase_ids: list[str]) -> dict | None:
+        raise NotImplementedError("Not implemented")
+
+    @abstractmethod
+    def insert(self, documents: list[dict], index_name: str, dataset_ids: str = None) -> list[str]:
+        raise NotImplementedError("Not implemented")
+
+    @abstractmethod
+    def update(self, condition: dict, new_value: dict, index_name: str, dataset_id: str) -> bool:
+        raise NotImplementedError("Not implemented")
+
+    def delete(self, condition: dict, index_name: str, dataset_id: str) -> int:
+        inf_conn = self.connPool.get_conn()
+        db_instance = inf_conn.get_database(self.dbName)
+        table_name = f"{index_name}_{dataset_id}"
+        try:
+            table_instance = db_instance.get_table(table_name)
+        except Exception:
+            self.logger.warning(f"Skipped deleting from table {table_name} since the table doesn't exist.")
+            return 0
+        filter = self.equivalent_condition_to_str(condition, table_instance)
+        self.logger.debug(f"INFINITY delete table {table_name}, filter {filter}.")
+        res = table_instance.delete(filter)
+        self.connPool.release_conn(inf_conn)
+        return res.deleted_rows
+
+    """
+    Helper functions for search result
+    """
+
+    def get_total(self, res: tuple[pd.DataFrame, int] | pd.DataFrame) -> int:
+        if isinstance(res, tuple):
+            return res[1]
+        return len(res)
+
+    def get_doc_ids(self, res: tuple[pd.DataFrame, int] | pd.DataFrame) -> list[str]:
+        if isinstance(res, tuple):
+            res = res[0]
+        return list(res["id"])
+
+    @abstractmethod
+    def get_fields(self, res: tuple[pd.DataFrame, int] | pd.DataFrame, fields: list[str]) -> dict[str, dict]:
+        raise NotImplementedError("Not implemented")
+
+    def get_highlight(self, res: tuple[pd.DataFrame, int] | pd.DataFrame, keywords: list[str], field_name: str):
+        if isinstance(res, tuple):
+            res = res[0]
+        ans = {}
+        num_rows = len(res)
+        column_id = res["id"]
+        if field_name not in res:
+            return {}
+        for i in range(num_rows):
+            id = column_id[i]
+            txt = res[field_name][i]
+            if re.search(r"<em>[^<>]+</em>", txt, flags=re.IGNORECASE | re.MULTILINE):
+                ans[id] = txt
+                continue
+            txt = re.sub(r"[\r\n]", " ", txt, flags=re.IGNORECASE | re.MULTILINE)
+            txt_list = []
+            for t in re.split(r"[.?!;\n]", txt):
+                if is_english([t]):
+                    for w in keywords:
+                        t = re.sub(
+                            r"(^|[ .?/'\"\(\)!,:;-])(%s)([ .?/'\"\(\)!,:;-])" % re.escape(w),
+                            r"\1<em>\2</em>\3",
+                            t,
+                            flags=re.IGNORECASE | re.MULTILINE,
+                        )
+                else:
+                    for w in sorted(keywords, key=len, reverse=True):
+                        t = re.sub(
+                            re.escape(w),
+                            f"<em>{w}</em>",
+                            t,
+                            flags=re.IGNORECASE | re.MULTILINE,
+                        )
+                if not re.search(r"<em>[^<>]+</em>", t, flags=re.IGNORECASE | re.MULTILINE):
+                    continue
+                txt_list.append(t)
+            if txt_list:
+                ans[id] = "...".join(txt_list)
+            else:
+                ans[id] = txt
+        return ans
+
+    def get_aggregation(self, res: tuple[pd.DataFrame, int] | pd.DataFrame, field_name: str):
+        """
+        Manual aggregation for tag fields since Infinity doesn't provide native aggregation
+        """
+        from collections import Counter
+
+        # Extract DataFrame from result
+        if isinstance(res, tuple):
+            df, _ = res
+        else:
+            df = res
+
+        if df.empty or field_name not in df.columns:
+            return []
+
+        # Aggregate tag counts
+        tag_counter = Counter()
+
+        for value in df[field_name]:
+            if pd.isna(value) or not value:
+                continue
+
+            # Handle different tag formats
+            if isinstance(value, str):
+                # Split by ### for tag_kwd field or comma for other formats
+                if field_name == "tag_kwd" and "###" in value:
+                    tags = [tag.strip() for tag in value.split("###") if tag.strip()]
+                else:
+                    # Try comma separation as fallback
+                    tags = [tag.strip() for tag in value.split(",") if tag.strip()]
+
+                for tag in tags:
+                    if tag:  # Only count non-empty tags
+                        tag_counter[tag] += 1
+            elif isinstance(value, list):
+                # Handle list format
+                for tag in value:
+                    if tag and isinstance(tag, str):
+                        tag_counter[tag.strip()] += 1
+
+        # Return as list of [tag, count] pairs, sorted by count descending
+        return [[tag, count] for tag, count in tag_counter.most_common()]
+
+    """
+    SQL
+    """
+
+    def sql(self, sql: str, fetch_size: int, format: str):
+        raise NotImplementedError("Not implemented")

common/doc_store/infinity_conn_pool.py

@@ -0,0 +1,85 @@
+#
+#  Copyright 2025 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 time
+
+import infinity
+from infinity.connection_pool import ConnectionPool
+from infinity.errors import ErrorCode
+
+from common import settings
+from common.decorator import singleton
+
+
+@singleton
+class InfinityConnectionPool:
+
+    def __init__(self):
+        if hasattr(settings, "INFINITY"):
+            self.INFINITY_CONFIG = settings.INFINITY
+        else:
+            self.INFINITY_CONFIG = settings.get_base_config("infinity", {"uri": "infinity:23817"})
+
+        infinity_uri = self.INFINITY_CONFIG["uri"]
+        if ":" in infinity_uri:
+            host, port = infinity_uri.split(":")
+            self.infinity_uri = infinity.common.NetworkAddress(host, int(port))
+
+        for _ in range(24):
+            try:
+                conn_pool = ConnectionPool(self.infinity_uri, max_size=4)
+                inf_conn = conn_pool.get_conn()
+                res = inf_conn.show_current_node()
+                if res.error_code == ErrorCode.OK and res.server_status in ["started", "alive"]:
+                    self.conn_pool = conn_pool
+                    conn_pool.release_conn(inf_conn)
+                    break
+            except Exception as e:
+                logging.warning(f"{str(e)}. Waiting Infinity {infinity_uri} to be healthy.")
+                time.sleep(5)
+
+        if self.conn_pool is None:
+            msg = f"Infinity {infinity_uri} is unhealthy in 120s."
+            logging.error(msg)
+            raise Exception(msg)
+
+        logging.info(f"Infinity {infinity_uri} is healthy.")
+
+    def get_conn_pool(self):
+        return self.conn_pool
+
+    def refresh_conn_pool(self):
+        try:
+            inf_conn = self.conn_pool.get_conn()
+            res = inf_conn.show_current_node()
+            if res.error_code == ErrorCode.OK and res.server_status in ["started", "alive"]:
+                return self.conn_pool
+            else:
+                raise Exception(f"{res.error_code}: {res.server_status}")
+
+        except Exception as e:
+            logging.error(str(e))
+            if hasattr(self, "conn_pool") and self.conn_pool:
+                self.conn_pool.destroy()
+                self.conn_pool = ConnectionPool(self.infinity_uri, max_size=32)
+                return self.conn_pool
+
+    def __del__(self):
+        if hasattr(self, "conn_pool") and self.conn_pool:
+            self.conn_pool.destroy()
+
+
+INFINITY_CONN = InfinityConnectionPool()

common/metadata_utils.py

@@ -13,6 +13,7 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 #
+import ast
 import logging
 from typing import Any, Callable, Dict
 
@@ -44,36 +45,55 @@ def meta_filter(metas: dict, filters: list[dict], logic: str = "and"):
     def filter_out(v2docs, operator, value):
         ids = []
         for input, docids in v2docs.items():
+
             if operator in ["=", "≠", ">", "<", "≥", "≤"]:
                 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 == "in", str(input).lower() in str(value).lower()),
-                (operator == "not in", str(input).lower() not in str(value).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
+                    if isinstance(input, list):
+                        input = input[0]
+                    input = ast.literal_eval(input)
+                    value = ast.literal_eval(value)
                 except Exception:
                     pass
+            if isinstance(input, str):
+                input = input.lower()
+            if isinstance(value, str):
+                value = value.lower()
+
+            matched = False
+            try:
+                if operator == "contains":
+                    matched = input in value if not isinstance(input, list) else all(i in value for i in input)
+                elif operator == "not contains":
+                    matched = input not in value if not isinstance(input, list) else all(i not in value for i in input)
+                elif operator == "in":
+                    matched = input in value if not isinstance(input, list) else all(i in value for i in input)
+                elif operator == "not in":
+                    matched = input not in value if not isinstance(input, list) else all(i not in value for i in input)
+                elif operator == "start with":
+                    matched = str(input).lower().startswith(str(value).lower()) if not isinstance(input, list) else "".join([str(i).lower() for i in input]).startswith(str(value).lower())
+                elif operator == "end with":
+                    matched = str(input).lower().endswith(str(value).lower()) if not isinstance(input, list) else "".join([str(i).lower() for i in input]).endswith(str(value).lower())
+                elif operator == "empty":
+                    matched = not input
+                elif operator == "not empty":
+                    matched = bool(input)
+                elif operator == "=":
+                    matched = input == value
+                elif operator == "≠":
+                    matched = input != value
+                elif operator == ">":
+                    matched = input > value
+                elif operator == "<":
+                    matched = input < value
+                elif operator == "≥":
+                    matched = input >= value
+                elif operator == "≤":
+                    matched = input <= value
+            except Exception:
+                pass
+
+            if matched:
+                ids.extend(docids)
         return ids
 
     for k, v2docs in metas.items():
@@ -145,6 +165,18 @@ async def apply_meta_data_filter(
     return doc_ids
 
 
+def dedupe_list(values: list) -> list:
+    seen = set()
+    deduped = []
+    for item in values:
+        key = str(item)
+        if key in seen:
+            continue
+        seen.add(key)
+        deduped.append(item)
+    return deduped
+
+
 def update_metadata_to(metadata, meta):
     if not meta:
         return metadata
@@ -156,11 +188,13 @@ def update_metadata_to(metadata, meta):
             return metadata
     if not isinstance(meta, dict):
         return metadata
+
     for k, v in meta.items():
         if isinstance(v, list):
             v = [vv for vv in v if isinstance(vv, str)]
             if not v:
                 continue
+            v = dedupe_list(v)
         if not isinstance(v, list) and not isinstance(v, str):
             continue
         if k not in metadata:
@@ -171,6 +205,7 @@ def update_metadata_to(metadata, meta):
                 metadata[k].extend(v)
             else:
                 metadata[k].append(v)
+            metadata[k] = dedupe_list(metadata[k])
         else:
             metadata[k] = v
 
@@ -202,4 +237,4 @@ def metadata_schema(metadata: list|None) -> Dict[str, Any]:
     }
 
     json_schema["additionalProperties"] = False
-    return json_schema
+    return json_schema

common/query_base.py

@@ -0,0 +1,72 @@
+#
+#  Copyright 2025 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 re
+from abc import ABC, abstractmethod
+
+
+class QueryBase(ABC):
+
+    @staticmethod
+    def is_chinese(line):
+        arr = re.split(r"[ \t]+", line)
+        if len(arr) <= 3:
+            return True
+        e = 0
+        for t in arr:
+            if not re.match(r"[a-zA-Z]+$", t):
+                e += 1
+        return e * 1.0 / len(arr) >= 0.7
+
+    @staticmethod
+    def sub_special_char(line):
+        return re.sub(r"([:\{\}/\[\]\-\*\"\(\)\|\+~\^])", r"\\\1", line).strip()
+
+    @staticmethod
+    def rmWWW(txt):
+        patts = [
+            (
+                r"是*(怎么办|什么样的|哪家|一下|那家|请问|啥样|咋样了|什么时候|何时|何地|何人|是否|是不是|多少|哪里|怎么|哪儿|怎么样|如何|哪些|是啥|啥是|啊|吗|呢|吧|咋|什么|有没有|呀|谁|哪位|哪个)是*",
+                "",
+            ),
+            (r"(^| )(what|who|how|which|where|why)('re|'s)? ", " "),
+            (
+                r"(^| )('s|'re|is|are|were|was|do|does|did|don't|doesn't|didn't|has|have|be|there|you|me|your|my|mine|just|please|may|i|should|would|wouldn't|will|won't|done|go|for|with|so|the|a|an|by|i'm|it's|he's|she's|they|they're|you're|as|by|on|in|at|up|out|down|of|to|or|and|if) ",
+                " ")
+        ]
+        otxt = txt
+        for r, p in patts:
+            txt = re.sub(r, p, txt, flags=re.IGNORECASE)
+        if not txt:
+            txt = otxt
+        return txt
+
+    @staticmethod
+    def add_space_between_eng_zh(txt):
+        # (ENG/ENG+NUM) + ZH
+        txt = re.sub(r'([A-Za-z]+[0-9]+)([\u4e00-\u9fa5]+)', r'\1 \2', txt)
+        # ENG + ZH
+        txt = re.sub(r'([A-Za-z])([\u4e00-\u9fa5]+)', r'\1 \2', txt)
+        # ZH + (ENG/ENG+NUM)
+        txt = re.sub(r'([\u4e00-\u9fa5]+)([A-Za-z]+[0-9]+)', r'\1 \2', txt)
+        txt = re.sub(r'([\u4e00-\u9fa5]+)([A-Za-z])', r'\1 \2', txt)
+        return txt
+
+    @abstractmethod
+    def question(self, text, tbl, min_match):
+        """
+        Returns a query object based on the input text, table, and minimum match criteria.
+        """
+        raise NotImplementedError("Not implemented")

common/settings.py

@@ -39,6 +39,9 @@ from rag.utils.oss_conn import RAGFlowOSS
 
 from rag.nlp import search
 
+import memory.utils.es_conn as memory_es_conn
+import memory.utils.infinity_conn as memory_infinity_conn
+
 LLM = None
 LLM_FACTORY = None
 LLM_BASE_URL = None
@@ -79,6 +82,7 @@ DOC_ENGINE_INFINITY = (DOC_ENGINE.lower() == "infinity")
 
 
 docStoreConn = None
+msgStoreConn = None
 
 retriever = None
 kg_retriever = None
@@ -166,7 +170,7 @@ def init_settings():
     global DATABASE_TYPE, DATABASE
     DATABASE_TYPE = os.getenv("DB_TYPE", "mysql")
     DATABASE = decrypt_database_config(name=DATABASE_TYPE)
-
+    
     global ALLOWED_LLM_FACTORIES, LLM_FACTORY, LLM_BASE_URL
     llm_settings = get_base_config("user_default_llm", {}) or {}
     llm_default_models = llm_settings.get("default_models", {}) or {}
@@ -256,6 +260,15 @@ def init_settings():
     else:
         raise Exception(f"Not supported doc engine: {DOC_ENGINE}")
 
+    global msgStoreConn
+    # use the same engine for message store
+    if DOC_ENGINE == "elasticsearch":
+        ES = get_base_config("es", {})
+        msgStoreConn = memory_es_conn.ESConnection()
+    elif DOC_ENGINE == "infinity":
+        INFINITY = get_base_config("infinity", {"uri": "infinity:23817"})
+        msgStoreConn = memory_infinity_conn.InfinityConnection()
+
     global AZURE, S3, MINIO, OSS, GCS
     if STORAGE_IMPL_TYPE in ['AZURE_SPN', 'AZURE_SAS']:
         AZURE = get_base_config("azure", {})
@@ -321,6 +334,9 @@ def init_settings():
     DOC_BULK_SIZE = int(os.environ.get("DOC_BULK_SIZE", 4))
     EMBEDDING_BATCH_SIZE = int(os.environ.get("EMBEDDING_BATCH_SIZE", 16))
 
+    os.environ["DOTNET_SYSTEM_GLOBALIZATION_INVARIANT"] = "1"
+
+
 def check_and_install_torch():
     global PARALLEL_DEVICES
     try:

common/time_utils.py

@@ -14,6 +14,7 @@
 #  limitations under the License.
 
 import datetime
+import logging
 import time
 
 def current_timestamp():
@@ -123,4 +124,31 @@ def delta_seconds(date_string: str):
         3600.0  # If current time is 2024-01-01 13:00:00
     """
     dt = datetime.datetime.strptime(date_string, "%Y-%m-%d %H:%M:%S")
-    return (datetime.datetime.now() - dt).total_seconds()
+    return (datetime.datetime.now() - dt).total_seconds()
+
+
+def format_iso_8601_to_ymd_hms(time_str: str) -> str:
+    """
+    Convert ISO 8601 formatted string to "YYYY-MM-DD HH:MM:SS" format.
+
+    Args:
+        time_str: ISO 8601 date string (e.g. "2024-01-01T12:00:00Z")
+
+    Returns:
+        str: Date string in "YYYY-MM-DD HH:MM:SS" format
+
+    Example:
+        >>> format_iso_8601_to_ymd_hms("2024-01-01T12:00:00Z")
+        '2024-01-01 12:00:00'
+    """
+    from dateutil import parser
+
+    try:
+        if parser.isoparse(time_str):
+            dt = datetime.datetime.fromisoformat(time_str.replace("Z", "+00:00"))
+            return dt.strftime("%Y-%m-%d %H:%M:%S")
+        else:
+            return time_str
+    except Exception as e:
+        logging.error(str(e))
+        return time_str

conf/llm_factories.json

@@ -1258,6 +1258,12 @@
             "status": "1",
             "rank": "810",
             "llm": [
+                {
+                    "llm_name": "MiniMax-M2.1",
+                    "tags": "LLM,CHAT,200k",
+                    "max_tokens": 200000,
+                    "model_type": "chat"
+                },
                 {
                     "llm_name": "MiniMax-M2",
                     "tags": "LLM,CHAT,200k",

conf/message_infinity_mapping.json

@@ -0,0 +1,19 @@
+{
+  "id": {"type": "varchar", "default": ""},
+  "message_id": {"type": "integer", "default": 0},
+  "message_type_kwd": {"type": "varchar", "default": ""},
+  "source_id": {"type":  "integer", "default": 0},
+  "memory_id": {"type": "varchar", "default": ""},
+  "user_id": {"type": "varchar", "default": ""},
+  "agent_id": {"type": "varchar", "default": ""},
+  "session_id": {"type": "varchar", "default": ""},
+  "valid_at": {"type":  "varchar", "default": ""},
+  "valid_at_flt": {"type": "float", "default": 0.0},
+  "invalid_at": {"type":  "varchar", "default": ""},
+  "invalid_at_flt": {"type": "float", "default": 0.0},
+  "forget_at": {"type":  "varchar", "default": ""},
+  "forget_at_flt": {"type": "float", "default": 0.0},
+  "status_int": {"type": "integer", "default": 1},
+  "zone_id": {"type": "integer", "default": 0},
+  "content": {"type": "varchar", "default": "", "analyzer": ["rag-coarse", "rag-fine"], "comment": "content_ltks"}
+}

deepdoc/parser/docling_parser.py

@@ -78,14 +78,21 @@ class DoclingParser(RAGFlowPdfParser):
     def __images__(self, fnm, zoomin: int = 1, page_from=0, page_to=600, callback=None):
         self.page_from = page_from
         self.page_to = page_to
+        bytes_io = None
         try:
-            opener = pdfplumber.open(fnm) if isinstance(fnm, (str, PathLike)) else pdfplumber.open(BytesIO(fnm))
+            if not isinstance(fnm, (str, PathLike)):
+                bytes_io = BytesIO(fnm)
+
+            opener = pdfplumber.open(fnm) if isinstance(fnm, (str, PathLike)) else pdfplumber.open(bytes_io)
             with opener as pdf:
                 pages = pdf.pages[page_from:page_to]
                 self.page_images = [p.to_image(resolution=72 * zoomin, antialias=True).original for p in pages]
         except Exception as e:
             self.page_images = []
             self.logger.exception(e)
+        finally:
+            if bytes_io:
+                bytes_io.close()
 
     def _make_line_tag(self,bbox: _BBox) -> str:
         if bbox is None:

deepdoc/parser/pdf_parser.py

@@ -476,11 +476,13 @@ class RAGFlowPdfParser:
         self.boxes = bxs
 
     def _naive_vertical_merge(self, zoomin=3):
-        bxs = self._assign_column(self.boxes, zoomin)
+        #bxs = self._assign_column(self.boxes, zoomin)
+        bxs = self.boxes
 
         grouped = defaultdict(list)
         for b in bxs:
-            grouped[(b["page_number"], b.get("col_id", 0))].append(b)
+            # grouped[(b["page_number"], b.get("col_id", 0))].append(b)
+            grouped[(b["page_number"], "x")].append(b)
 
         merged_boxes = []
         for (pg, col), bxs in grouped.items():
@@ -551,7 +553,7 @@ class RAGFlowPdfParser:
 
             merged_boxes.extend(bxs)
 
-        self.boxes = sorted(merged_boxes, key=lambda x: (x["page_number"], x.get("col_id", 0), x["top"]))
+        #self.boxes = sorted(merged_boxes, key=lambda x: (x["page_number"], x.get("col_id", 0), x["top"]))
 
     def _final_reading_order_merge(self, zoomin=3):
         if not self.boxes:
@@ -1061,8 +1063,8 @@ class RAGFlowPdfParser:
 
                     self.total_page = len(self.pdf.pages)
 
-        except Exception:
-            logging.exception("RAGFlowPdfParser __images__")
+        except Exception as e:
+            logging.exception(f"RAGFlowPdfParser __images__, exception: {e}")
         logging.info(f"__images__ dedupe_chars cost {timer() - start}s")
 
         self.outlines = []
@@ -1447,6 +1449,7 @@ class VisionParser(RAGFlowPdfParser):
     def __init__(self, vision_model, *args, **kwargs):
         super().__init__(*args, **kwargs)
         self.vision_model = vision_model
+        self.outlines = []
 
     def __images__(self, fnm, zoomin=3, page_from=0, page_to=299, callback=None):
         try:

deepdoc/vision/operators.py

@@ -16,6 +16,7 @@
 
 import logging
 import sys
+import ast
 import six
 import cv2
 import numpy as np
@@ -108,7 +109,14 @@ class NormalizeImage:
 
     def __init__(self, scale=None, mean=None, std=None, order='chw', **kwargs):
         if isinstance(scale, str):
-            scale = eval(scale)
+            try:
+                scale = float(scale)
+            except ValueError:
+                if '/' in scale:
+                    parts = scale.split('/')
+                    scale = ast.literal_eval(parts[0]) / ast.literal_eval(parts[1])
+                else:
+                    scale = ast.literal_eval(scale)
         self.scale = np.float32(scale if scale is not None else 1.0 / 255.0)
         mean = mean if mean is not None else [0.485, 0.456, 0.406]
         std = std if std is not None else [0.229, 0.224, 0.225]

docker/.env

@@ -1,3 +1,10 @@
+# -----------------------------------------------------------------------------
+# SECURITY WARNING: DO NOT DEPLOY WITH DEFAULT PASSWORDS
+# For non-local deployments, please change all passwords (ELASTIC_PASSWORD, 
+# MYSQL_PASSWORD, MINIO_PASSWORD, etc.) to strong, unique values.
+# You can generate a random string using: openssl rand -hex 32
+# -----------------------------------------------------------------------------
+
 # ------------------------------
 # docker env var for specifying vector db type at startup
 # (based on the vector db type, the corresponding docker
@@ -30,6 +37,7 @@ ES_HOST=es01
 ES_PORT=1200
 
 # The password for Elasticsearch.
+# WARNING: Change this for production!
 ELASTIC_PASSWORD=infini_rag_flow
 
 # the hostname where OpenSearch service is exposed, set it not the same as elasticsearch
@@ -85,6 +93,7 @@ OB_DATAFILE_SIZE=${OB_DATAFILE_SIZE:-20G}
 OB_LOG_DISK_SIZE=${OB_LOG_DISK_SIZE:-20G}
 
 # The password for MySQL.
+# WARNING: Change this for production!
 MYSQL_PASSWORD=infini_rag_flow
 # The hostname where the MySQL service is exposed
 MYSQL_HOST=mysql
@@ -128,11 +137,11 @@ ADMIN_SVR_HTTP_PORT=9381
 SVR_MCP_PORT=9382
 
 # The RAGFlow Docker image to download. v0.22+ doesn't include embedding models.
-RAGFLOW_IMAGE=infiniflow/ragflow:v0.22.1
+RAGFLOW_IMAGE=infiniflow/ragflow:v0.23.0
 
 # If you cannot download the RAGFlow Docker image:
-# RAGFLOW_IMAGE=swr.cn-north-4.myhuaweicloud.com/infiniflow/ragflow:v0.22.1
-# RAGFLOW_IMAGE=registry.cn-hangzhou.aliyuncs.com/infiniflow/ragflow:v0.22.1
+# RAGFLOW_IMAGE=swr.cn-north-4.myhuaweicloud.com/infiniflow/ragflow:v0.23.0
+# RAGFLOW_IMAGE=registry.cn-hangzhou.aliyuncs.com/infiniflow/ragflow:v0.23.0
 #
 # - For the `nightly` edition, uncomment either of the following:
 # RAGFLOW_IMAGE=swr.cn-north-4.myhuaweicloud.com/infiniflow/ragflow:nightly
@@ -234,9 +243,8 @@ REGISTER_ENABLED=1
 USE_DOCLING=false
 
 # Enable Mineru
-USE_MINERU=false
-MINERU_EXECUTABLE="$HOME/uv_tools/.venv/bin/mineru"
 # Uncommenting these lines will automatically add MinerU to the model provider whenever possible.
+# More details see https://ragflow.io/docs/faq#how-to-use-mineru-to-parse-pdf-documents.
 # MINERU_DELETE_OUTPUT=0   # keep output directory
 # MINERU_BACKEND=pipeline  # or another backend you prefer
 

docker/README.md

@@ -77,7 +77,7 @@ The [.env](./.env) file contains important environment variables for Docker.
 - `SVR_HTTP_PORT`  
   The port used to expose RAGFlow's HTTP API service to the host machine, allowing **external** access to the service running inside the Docker container. Defaults to `9380`.
 - `RAGFLOW-IMAGE`  
-  The Docker image edition. Defaults to `infiniflow/ragflow:v0.22.1`. The RAGFlow Docker image does not include embedding models.
+  The Docker image edition. Defaults to `infiniflow/ragflow:v0.23.0`. The RAGFlow Docker image does not include embedding models.
 
   
 > [!TIP]  

docker/docker-compose-base.yml

@@ -72,7 +72,7 @@ services:
   infinity:
     profiles:
       - infinity
-    image: infiniflow/infinity:v0.6.11
+    image: infiniflow/infinity:v0.6.15
     volumes:
       - infinity_data:/var/infinity
       - ./infinity_conf.toml:/infinity_conf.toml

docker/infinity_conf.toml

@@ -1,5 +1,5 @@
 [general]
-version                  = "0.6.11"
+version                  = "0.6.15"
 time_zone                = "utc-8"
 
 [network]

docs/basics/_category_.json

@@ -0,0 +1,8 @@
+{
+  "label": "Basics",
+  "position": 2,
+  "link": {
+    "type": "generated-index",
+    "description": "Basic concepts."
+  }
+}

docs/basics/agent_context_engine.md

@@ -0,0 +1,61 @@
+---
+sidebar_position: 2
+slug: /what_is_agent_context_engine
+---
+
+# What is Agent context engine?
+
+From 2025, a silent revolution began beneath the dazzling surface of AI Agents. While the world marveled at agents that could write code, analyze data, and automate workflows, a fundamental bottleneck emerged: why do even the most advanced agents still stumble on simple questions, forget previous conversations, or misuse available tools?
+
+The answer lies not in the intelligence of the Large Language Model (LLM) itself, but in the quality of the Context it receives. An LLM, no matter how powerful, is only as good as the information we feed it. Today’s cutting-edge agents are often crippled by a cumbersome, manual, and error-prone process of context assembly—a process known as Context Engineering.
+
+This is where the Agent Context Engine comes in. It is not merely an incremental improvement but a foundational shift, representing the evolution of RAG from a singular technique into the core data and intelligence substrate for the entire Agent ecosystem.
+
+## Beyond the hype: The reality of today's "intelligent" Agents
+Today, the “intelligence” behind most AI Agents hides a mountain of human labor. Developers must:
+
+- Hand-craft elaborate prompt templates
+- Hard-code document-retrieval logic for every task
+- Juggle tool descriptions, conversation history, and knowledge snippets inside a tiny context window
+- Repeat the whole process for each new scenario
+
+This pattern is called Context Engineering. It is deeply tied to expert know-how, almost impossible to scale, and prohibitively expensive to maintain. When an enterprise needs to keep dozens of distinct agents alive, the artisanal workshop model collapses under its own weight.
+
+The mission of an Agent Context Engine is to turn Context Engineering from an “art” into an industrial-grade science.
+
+Deconstructing the Agent Context Engine
+So, what exactly is an Agent Context Engine? It is a unified, intelligent, and automated platform responsible for the end-to-end process of assembling the optimal context for an LLM or Agent at the moment of inference. It moves from artisanal crafting to industrialized production.
+At its core, an Agent Context Engine is built on a triumvirate of next-generation retrieval capabilities, seamlessly integrated into a single service layer:
+
+1. The Knowledge Core (Advanced RAG): This is the evolution of traditional RAG. It moves beyond simple chunk-and-embed to intelligently process static, private enterprise knowledge. Techniques like TreeRAG (building LLM-generated document outlines for "locate-then-expand" retrieval) and GraphRAG (extracting entity networks to find semantically distant connections) work to close the "semantic gap." The engine’s Ingestion Pipeline acts as the ETL for unstructured data, parsing multi-format documents and using LLMs to enrich content with summaries, metadata, and structure before indexing.
+
+2. The Memory Layer: An Agent’s intelligence is defined by its ability to learn from interaction. The Memory Layer is a specialized retrieval system for dynamic, episodic data: conversation history, user preferences, and the agent’s own internal state (e.g., "waiting for human input"). It manages the lifecycle of this data—storing raw dialogue, triggering summarization into semantic memory, and retrieving relevant past interactions to provide continuity and personalization. Technologically, it is a close sibling to RAG, but focused on a temporal stream of data.
+
+3. The Tool Orchestrator: As MCP (Model Context Protocol) enables the connection of hundreds of internal services as tools, a new problem arises: tool selection. The Context Engine solves this with Tool Retrieval. Instead of dumping all tool descriptions into the prompt, it maintains an index of tools and—critically—an index of Playbooks or Guidelines (best practices on when and how to use tools). For a given task, it retrieves only the most relevant tools and instructions, transforming the LLM’s job from "searching a haystack" to "following a recipe."
+
+## Why we need a dedicated engine? The case for a unified substrate
+
+The necessity of an Agent Context Engine becomes clear when we examine the alternative: siloed, manually wired components.
+
+- The Data Silo Problem: Knowledge, memory, and tools reside in separate systems, requiring complex integration for each new agent.
+- The Assembly Line Bottleneck: Developers spend more time on context plumbing than on agent logic, slowing innovation to a crawl.
+- The "Context Ownership" Dilemma: In manually engineered systems, context logic is buried in code, owned by developers, and opaque to business users. An Engine makes context a configurable, observable, and customer-owned asset.
+
+The shift from Context Engineering to a Context Platform/Engine marks the maturation of enterprise AI, as summarized in the table below:
+
+| Dimension           | Context engineering (present)                                              | Context engineering/Platform (future)                                                               |
+| ------------------- | -------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------- |
+| Context creation    | Manual, artisanal work by developers and prompt engineers.                 | Automated, driven by intelligent ingestion pipelines and configurable rules.                        |
+| Context delivery    | Hard-coded prompts and static retrieval logic embedded in agent workflows. | Dynamic, real-time retrieval and assembly based on the agent's live state and intent.               |
+| Context maintenance | A development and operational burden, logic locked in code.                | A manageable platform function, with visibility and control returned to the business.               |
+
+
+## RAGFlow: A resolute march toward the context engine of Agents
+
+This is the future RAGFlow is forging.
+
+We left behind the label of “yet another RAG system” long ago. From DeepDoc—our deeply-optimized, multimodal document parser—to the bleeding-edge architectures that bridge semantic chasms in complex RAG scenarios, all the way to a full-blown, enterprise-grade ingestion pipeline, every evolutionary step RAGFlow takes is a deliberate stride toward the ultimate form: an Agentic Context Engine.
+
+We believe tomorrow’s enterprise AI advantage will hinge not on who owns the largest model, but on who can feed that model the highest-quality, most real-time, and most relevant context. An Agentic Context Engine is the critical infrastructure that turns this vision into reality.
+
+In the paradigm shift from “hand-crafted prompts” to “intelligent context,” RAGFlow is determined to be the most steadfast propeller and enabler. We invite every developer, enterprise, and researcher who cares about the future of AI agents to follow RAGFlow’s journey—so together we can witness and build the cornerstone of the next-generation AI stack.

docs/basics/rag.md

@@ -0,0 +1,107 @@
+---
+sidebar_position: 1
+slug: /what_is_rag
+---
+
+# What is Retreival-Augmented-Generation (RAG)？
+
+Since large language models (LLMs) became the focus of technology, their ability to handle general knowledge has been astonishing. However, when questions shift to internal corporate documents, proprietary knowledge bases, or real-time data, the limitations of LLMs become glaringly apparent: they cannot access private information outside their training data. Retrieval-Augmented Generation (RAG) was born precisely to address this core need. Before an LLM generates an answer, it first retrieves the most relevant context from an external knowledge base and inputs it as "reference material" to the LLM, thereby guiding it to produce accurate answers. In short, RAG elevates LLMs from "relying on memory" to "having evidence to rely on," significantly improving their accuracy and trustworthiness in specialized fields and real-time information queries.
+
+## Why RAG is important?
+
+Although LLMs excel in language understanding and generation, they have inherent limitations:
+
+- Static Knowledge: The model's knowledge is based on a data snapshot from its training time and cannot be automatically updated, making it difficult to perceive the latest information.
+- Blind Spot to External Data: They cannot directly access corporate private documents, real-time information streams, or domain-specific content.
+- Hallucination Risk: When lacking accurate evidence, they may still fabricate plausible-sounding but false answers to maintain conversational fluency.
+
+The introduction of RAG provides LLMs with real-time, credible "factual grounding." Its core mechanism is divided into two stages:
+
+- Retrieval Stage: Based on the user's question, quickly retrieve the most relevant documents or data fragments from an external knowledge base.
+- Generation Stage: The LLM organizes and generates the final answer by incorporating the retrieved information as context, combined with its own linguistic capabilities.
+
+This upgrades LLMs from "speaking from memory" to "speaking with documentation," significantly enhancing reliability in professional and enterprise-level applications.
+
+## How RAG works?
+
+Retrieval-Augmented Generation enables LLMs to generate higher-quality responses by leveraging real-time, external, or private data sources through the introduction of an information retrieval mechanism. Its workflow can be divided into following key steps:
+
+### Data processing and vectorization
+
+The knowledge required by RAG comes from unstructured data in various formats, such as documents, database records, or API return content. This data typically needs to be chunked, then transformed into vectors via an embedding model, and stored in a vector database.
+
+Why is Chunking Needed? Indexing entire documents directly faces the following problems:
+
+- Decreased Retrieval Precision: Vectorizing long documents leads to semantic "averaging," losing details.
+- Context Length Limitation: LLMs have a finite context window, requiring filtering of the most relevant parts for input.
+- Cost and Efficiency: Embedding computation and retrieval costs are higher for long texts.
+
+Therefore, an intelligent chunking strategy is key to balancing information integrity, retrieval granularity, and computational efficiency.
+
+### Retrieve relevant information
+
+The user's query is also converted into a vector to perform semantic relevance searches (e.g., calculating cosine similarity) in the vector database, matching and recalling the most relevant text fragments.
+
+### Context construction and answer generation
+
+The retrieved relevant content is added to the LLM's context as factual grounding, and the LLM finally generates the answer. Therefore, RAG can be seen as Context Engineering 1.0 for automated context construction.
+
+## Deep dive into existing RAG architecture: beyond vector retrieval
+
+An industrial-grade RAG system is far from being as simple as "vector search + LLM"; its complexity and challenges are primarily embedded in the retrieval process.
+
+### Data complexity: multimodal document processing
+
+Core Challenge: Corporate knowledge mostly exists in the form of multimodal documents containing text, charts, tables, and formulas. Simple OCR extraction loses a large amount of semantic information.
+
+Advanced Practice: Leading solutions, such as RAGFlow, tend to use Visual Language Models (VLM) or specialized parsing models like DeepDoc to "translate" multimodal documents into unimodal text rich in structural and semantic information. Converting multimodal information into high-quality unimodal text has become standard practice for advanced RAG.
+
+### The complexity of chunking: the trade-off between precision and context
+
+A simple "chunk-embed-retrieve" pipeline has an inherent contradiction:
+- Semantic Matching requires small text chunks to ensure clear semantic focus.
+- Context Understanding requires large text chunks to ensure complete and coherent information.
+
+This forces system design into a difficult trade-off between "precise but fragmented" and "complete but vague."
+
+Advanced Practice: Leading solutions, such as RAGFlow, employ semantic enhancement techniques like constructing semantic tables of contents and knowledge graphs. These not only address semantic fragmentation caused by physical chunking but also enable the discovery of relevant content across documents based on entity-relationship networks.
+
+### Why is a vector database insufficient for serving RAG?
+
+Vector databases excel at semantic similarity search, but RAG requires precise and reliable answers, demanding more capabilities from the retrieval system:
+- Hybrid Search: Relying solely on vector retrieval may miss exact keyword matches (e.g., product codes, regulation numbers). Hybrid search, combining vector retrieval with keyword retrieval (BM25), ensures both semantic breadth and keyword precision.
+- Tensor or Multi-Vector Representation: To support cross-modal data, employing tensor or multi-vector representation has become an important trend.
+- Metadata Filtering: Filtering based on attributes like date, department, and type is a rigid requirement in business scenarios.
+
+Therefore, the retrieval layer of RAG is a composite system based on vector search but must integrate capabilities like full-text search, re-ranking, and metadata filtering.
+
+## RAG and memory: Retrieval from the same source but different streams
+
+Within the agent framework, the essence of the memory mechanism is the same as RAG: both retrieve relevant information from storage based on current needs. The key difference lies in the data source:
+- RAG: Targets pre-existing static or dynamic private data provided by the user in advance (e.g., documents, databases).
+- Memory: Targets dynamic data generated or perceived by the agent in real-time during interaction (e.g., conversation history, environmental state, tool execution results).
+They are highly consistent at the technical base (e.g., vector retrieval, keyword matching) and can be seen as the same retrieval capability applied in different scenarios ("existing knowledge" vs. "interaction memory"). A complete agent system often includes both an RAG module for inherent knowledge and a Memory module for interaction history.
+
+## RAG applications
+
+RAG has demonstrated clear value in several typical scenarios:
+
+1. Enterprise Knowledge Q&A and Internal Search  
+   By vectorizing corporate private data and combining it with an LLM, RAG can directly return natural language answers based on authoritative sources, rather than document lists. While meeting intelligent Q&A needs, it inherently aligns with corporate requirements for data security, access control, and compliance.
+2. Complex Document Understanding and Professional Q&A  
+   For structurally complex documents like contracts and regulations, the value of RAG lies in its ability to generate accurate, verifiable answers while maintaining context integrity. Its system accuracy largely depends on text chunking and semantic understanding strategies.
+3. Dynamic Knowledge Fusion and Decision Support  
+   In business scenarios requiring the synthesis of information from multiple sources, RAG evolves into a knowledge orchestration and reasoning support system for business decisions. Through a multi-path recall mechanism, it fuses knowledge from different systems and formats, maintaining factual consistency and logical controllability during the generation phase.
+
+## The future of RAG
+
+The evolution of RAG is unfolding along several clear paths:
+
+1. RAG as the data foundation for Agents  
+   RAG and agents have an architecture vs. scenario relationship. For agents to achieve autonomous and reliable decision-making and execution, they must rely on accurate and timely knowledge. RAG provides them with a standardized capability to access private domain knowledge and is an inevitable choice for building knowledge-aware agents.
+2. Advanced RAG: Using LLMs to optimize retrieval itself  
+   The core feature of next-generation RAG is fully utilizing the reasoning capabilities of LLMs to optimize the retrieval process, such as rewriting queries, summarizing or fusing results, or implementing intelligent routing. Empowering every aspect of retrieval with LLMs is key to breaking through current performance bottlenecks.
+3. Towards context engineering 2.0  
+   Current RAG can be viewed as Context Engineering 1.0, whose core is assembling static knowledge context for single Q&A tasks. The forthcoming Context Engineering 2.0 will extend with RAG technology at its core, becoming a system that automatically and dynamically assembles comprehensive context for agents. The context fused by this system will come not only from documents but also include interaction memory, available tools/skills, and real-time environmental information. This marks the transition of agent development from a "handicraft workshop" model to the industrial starting point of automated context engineering.
+
+The essence of RAG is to build a dedicated, efficient, and trustworthy external data interface for large language models; its core is Retrieval, not Generation. Starting from the practical need to solve private data access, its technical depth is reflected in the optimization of retrieval for complex unstructured data. With its deep integration into agent architectures and its development towards automated context engineering, RAG is evolving from a technology that improves Q&A quality into the core infrastructure for building the next generation of trustworthy, controllable, and scalable intelligent applications.

docs/configurations.md

@@ -99,7 +99,7 @@ RAGFlow utilizes MinIO as its object storage solution, leveraging its scalabilit
 - `SVR_HTTP_PORT`
   The port used to expose RAGFlow's HTTP API service to the host machine, allowing **external** access to the service running inside the Docker container. Defaults to `9380`.
 - `RAGFLOW-IMAGE`
-  The Docker image edition. Defaults to `infiniflow/ragflow:v0.22.1` (the RAGFlow Docker image without embedding models).
+  The Docker image edition. Defaults to `infiniflow/ragflow:v0.23.0` (the RAGFlow Docker image without embedding models).
 
 :::tip NOTE
 If you cannot download the RAGFlow Docker image, try the following mirrors.

docs/develop/build_docker_image.mdx

@@ -47,7 +47,7 @@ After building the infiniflow/ragflow:nightly image, you are ready to launch a f
 
 1. Edit Docker Compose Configuration
 
-Open the `docker/.env` file. Find the `RAGFLOW_IMAGE` setting and change the image reference from `infiniflow/ragflow:v0.22.1` to `infiniflow/ragflow:nightly` to use the pre-built image.
+Open the `docker/.env` file. Find the `RAGFLOW_IMAGE` setting and change the image reference from `infiniflow/ragflow:v0.23.0` to `infiniflow/ragflow:nightly` to use the pre-built image.
 
 
 2. Launch the Service

docs/guides/agent/agent_component_reference/http.md

@@ -0,0 +1,90 @@
+---
+sidebar_position: 30
+slug: /http_request_component
+---
+
+# HTTP request component
+
+A component that calls remote services. 
+
+---
+
+An **HTTP request** component lets you access remote APIs or services by providing a URL and an HTTP method, and then receive the response. You can customize headers, parameters, proxies, and timeout settings, and use common methods like GET and POST. It’s useful for exchanging data with external systems in a workflow.
+
+## Prerequisites
+
+- An accessible remote API or service.
+- Add a Token or credentials to the request header, if the target service requires authentication.
+
+## Configurations
+
+### Url
+
+*Required*. The complete request address, for example: http://api.example.com/data.
+
+### Method
+
+The HTTP request method to select. Available options:
+
+- GET
+- POST
+- PUT
+
+### Timeout
+
+The maximum waiting time for the request, in seconds. Defaults to `60`.
+
+### Headers
+
+Custom HTTP headers can be set here, for example:
+
+```http
+{
+  "Accept": "application/json",
+  "Cache-Control": "no-cache",
+  "Connection": "keep-alive"
+}
+```
+
+### Proxy
+
+Optional. The proxy server address to use for this request.
+
+### Clean HTML
+
+`Boolean`: Whether to remove HTML tags from the returned results and keep plain text only.
+
+### Parameter
+
+*Optional*. Parameters to send with the HTTP request. Supports key-value pairs:
+
+- To assign a value using a dynamic system variable, set it as Variable.
+- To override these dynamic values under certain conditions and use a fixed static value instead, Value is the appropriate choice.
+
+
+:::tip NOTE
+- For GET requests, these parameters are appended to the end of the URL.
+- For POST/PUT requests, they are sent as the request body.
+:::
+
+#### Example setting
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/http_settings.png)
+
+#### Example response
+
+```html
+{ "args": { "App": "RAGFlow", "Query": "How to do?", "Userid": "241ed25a8e1011f0b979424ebc5b108b" }, "headers": { "Accept": "/", "Accept-Encoding": "gzip, deflate, br, zstd", "Cache-Control": "no-cache", "Host": "httpbin.org", "User-Agent": "python-requests/2.32.2", "X-Amzn-Trace-Id": "Root=1-68c9210c-5aab9088580c130a2f065523" }, "origin": "185.36.193.38", "url": "https://httpbin.org/get?Userid=241ed25a8e1011f0b979424ebc5b108b&App=RAGFlow&Query=How+to+do%3F" }
+```
+
+### Output
+
+The global variable name for the output of the HTTP request component, which can be referenced by other components in the workflow.
+
+- `Result`: `string` The response returned by the remote service.
+
+## Example
+
+This is a usage example: a workflow sends a GET request from the **Begin** component to `https://httpbin.org/get` via the **HTTP Request_0** component, passes parameters to the server, and finally outputs the result through the **Message_0** component.
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/http_usage.PNG)

docs/guides/dataset/auto_metadata.md

@@ -0,0 +1,48 @@
+---
+sidebar_position: -6
+slug: /auto_metadata
+---
+
+# Auto-extract metadata
+
+Automatically extract metadata from uploaded files.
+
+---
+
+RAGFlow v0.23.0 introduces the Auto-metadata feature, which uses large language models to automatically extract and generate metadata for files—eliminating the need for manual entry. In a typical RAG pipeline, metadata serves two key purposes:
+
+- During the retrieval stage: Filters out irrelevant documents, narrowing the search scope to improve retrieval accuracy.
+- During the generation stage: If a text chunk is retrieved, its associated metadata is also passed to the LLM, providing richer contextual information about the source document to aid answer generation.
+
+
+:::danger WARNING
+Enabling TOC extraction requires significant memory, computational resources, and tokens.
+:::
+
+
+
+## Procedure
+
+1. On your dataset's **Configuration** page, select an indexing model, which will be used to generate the knowledge graph, RAPTOR, auto-metadata, auto-keyword, and auto-question features for this dataset.
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/indexing_model.png)
+
+
+2. Click **Auto metadata** **>** **Settings** to go to the configuration page for automatic metadata generation rules.
+
+   _The configuration page for rules on automatically generating metadata appears._
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/auto_metadata_settings.png)
+
+3. Click **+** to add new fields and enter the configuration page.
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/metadata_field_settings.png)
+
+4. Enter a field name, such as Author, and add a description and examples in the Description section. This provides context to the large language model (LLM) for more accurate value extraction. If left blank, the LLM will extract values based only on the field name.
+
+5. To restrict the LLM to generating metadata from a predefined list, enable the Restrict to defined values mode and manually add the allowed values. The LLM will then only generate results from this preset range.
+
+6. Once configured, turn on the Auto-metadata switch on the Configuration page. All newly uploaded files will have these rules applied during parsing. For files that have already been processed, you must re-parse them to trigger metadata generation. You can then use the filter function to check the metadata generation status of your files.
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/enable_auto_metadata.png)
+

docs/guides/dataset/configure_child_chunking_strategy.md

@@ -0,0 +1,34 @@
+---
+sidebar_position: -4
+slug: /configure_child_chunking_strategy
+---
+
+# Configure child chunking strategy
+
+Set parent-child chunking strategy to improve retrieval.
+
+---
+
+A persistent challenge in practical RAG applications lies in a structural tension within the traditional "chunk-embed-retrieve" pipeline: a single text chunk is tasked with both semantic matching (recall) and contextual understanding (utilization)—two inherently conflicting objectives. Recall demands fine-grained, precise chunks, while answer generation requires coherent, informationally complete context.
+
+To resolve this tension, RAGFlow previously introduced the Table of Contents (TOC) enhancement feature, which uses a large language model (LLM) to generate document structure and automatically supplements missing context during retrieval based on that TOC. In version 0.23.0, this capability has been systematically integrated into the Ingestion Pipeline, and a novel parent-child chunking mechanism has been introduced.
+
+Under this mechanism, a document is first segmented into larger parent chunks, each maintaining a relatively complete semantic unit to ensure logical and background integrity. Each parent chunk can then be further subdivided into multiple child chunks for precise recall. During retrieval, the system first locates the most relevant text segments based on the child chunks while automatically associating and recalling their parent chunk. This approach maintains high recall relevance while providing ample semantic background for the generation phase.
+
+For instance, when processing a *Compliance Handbook*, a user query about "liability for breach" might precisely retrieve a child chunk stating, "The penalty for breach is 20% of the total contract value," but without context, it cannot clarify whether this clause applies to "minor breach" or "material breach." Leveraging the parent-child chunking mechanism, the system returns this child chunk along with its parent chunk, which contains the complete section of the clause. This allows the LLM to make accurate judgments based on broader context, avoiding misinterpretation.
+
+Through this dual-layer structure of "precise localization + contextual supplementation," RAGFlow ensures retrieval accuracy while significantly enhancing the reliability and completeness of generated answers.
+
+
+## Procedure
+
+1. On your dataset's **Configuration** page, find the **Child chunk are used for retrieval** toggle:
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/child_chunking.png)
+
+
+2. Set the delimiter for child chunks.
+
+3. This configuration applies to the **Chunker** component when it comes to ingestion pipeline settings:
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/child_chunking_parser.png)

docs/guides/dataset/configure_knowledge_base.md

@@ -133,7 +133,7 @@ See [Run retrieval test](./run_retrieval_test.md) for details.
 
 ## Search for dataset
 
-As of RAGFlow v0.22.1, the search feature is still in a rudimentary form, supporting only dataset search by name.
+As of RAGFlow v0.23.0, the search feature is still in a rudimentary form, supporting only dataset search by name.
 
 ![search dataset](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/search_datasets.jpg)
 

docs/guides/dataset/manage_metadata.md

@@ -0,0 +1,47 @@
+---
+sidebar_position: -5
+slug: /manage_metadata
+---
+
+# Manage metadata
+
+Manage metadata for your dataset and for your individual documents.
+
+---
+
+From v0.23.0 onwards, RAGFlow allows you to manage metadata both at the dataset level and for individual files.
+
+
+## Procedure
+
+1. Click on **Metadata** within your dataset to access the **Manage Metadata** page.
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/click_metadata.png)
+
+
+2. On the **Manage Metadata** page, you can do either of the following: 
+   - Edit Values: You can modify existing values. If you rename two values to be identical, they will be automatically merged.
+   - Delete: You can delete specific values or entire fields. These changes will apply to all associated files.
+
+   _The configuration page for rules on automatically generating metadata appears._
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/manage_metadata.png)
+
+3. To manage metadata for a single file, navigate to the file's details page as shown below. Click on the parsing method (e.g., **General**), then select **Set Metadata** to view or edit the file's metadata. Here, you can add, delete, or modify metadata fields for this specific file. Any edits made here will be reflected in the global statistics on the main Metadata management page for the knowledge base.
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/set_metadata.png)
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/edit_metadata.png)
+
+4. The filtering function operates at two levels: knowledge base management and retrieval. Within the dataset, click the Filter button to view the number of files associated with each value under existing metadata fields. By selecting specific values, you can display all linked files.
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/filter_metadata.png)
+
+5. Metadata filtering is also supported during the retrieval stage. In Chat, for example, you can set metadata filtering rules after configuring a knowledge base:
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/metadata_filtering_rules.png)
+
+-   **Automatic** Mode: The system automatically filters documents based on the user's query and the existing metadata in the knowledge base.
+-   **Semi-automatic** Mode: Users first define the filtering scope at the field level (e.g., for **Author**), and then the system automatically filters within that preset range.
+-   **Manual** Mode: Users manually set precise, value-specific filter conditions, supported by operators such as **Equals**, **Not equals**, **In**, **Not in**, and more.
+
+

docs/guides/dataset/select_pdf_parser.md

@@ -1,5 +1,5 @@
 ---
-sidebar_position: -4
+sidebar_position: -3
 slug: /select_pdf_parser
 ---
 

docs/guides/dataset/set_context_window.md

@@ -0,0 +1,25 @@
+---
+sidebar_position: -8
+slug: /set_context_window
+---
+
+# Set context window size
+
+Set context window size for images and tables to improve long-context RAG performances.
+
+---
+
+RAGFlow leverages built-in DeepDoc, along with external document models like MinerU and Docling, to parse document layouts. In previous versions, images and tables extracted based on document layout were treated as independent chunks. Consequently, if a search query did not directly match the content of an image or table, these elements would not be retrieved. However, real-world documents frequently interweave charts and tables with surrounding text, which often describes them. Therefore, recalling charts based on this contextual text is an essential capability.
+
+To address this, RAGFlow 0.23.0 introduces the **Image & table context window** feature. Inspired by key principles of the research-focused, open-source multimodal RAG project RAG-Anything, this functionality allows surrounding text and adjacent visuals to be grouped into a single chunk based on a user-configurable window size. This ensures they are retrieved together, significantly improving the recall accuracy for charts and tables.
+
+## Procedure
+
+1. On your dataset's **Configuration** page, find the **Image & table context window** slider:
+
+![](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/image_table_context_window.png)
+
+
+2. Adjust the number of context tokens according to your needs.
+
+   *The number in the red box indicates that approximately **N tokens** of text from above and below the image/table will be captured and inserted into the image or table chunk as contextual information. The capture process intelligently optimizes boundaries at punctuation marks to preserve semantic integrity. *

docs/guides/dataset/set_metadata.md

@@ -5,7 +5,7 @@ slug: /set_metadata
 
 # Set metadata
 
-Add metadata to an uploaded file
+Manually add metadata to an uploaded file
 
 ---
 
@@ -29,4 +29,4 @@ Ensure that your metadata is in JSON format; otherwise, your updates will not be
 
 ### Can I set metadata for multiple documents at once?
 
-No, you must set metadata *individually* for each document, as RAGFlow does not support batch setting of metadata. If you still consider this feature essential, please [raise an issue](https://github.com/infiniflow/ragflow/issues) explaining your use case and its importance.
+From v0.23.0 onwards, you can set metadata for each document individually or have the LLM auto-generate metadata for multiple files. See [Extract metadata](./auto_metadata.md) for details.

docs/guides/manage_files.md

@@ -87,4 +87,4 @@ RAGFlow's file management allows you to download an uploaded file:
 
 ![download_file](https://github.com/infiniflow/ragflow/assets/93570324/cf3b297f-7d9b-4522-bf5f-4f45743e4ed5)
 
-> As of RAGFlow v0.22.1, bulk download is not supported, nor can you download an entire folder. 
+> As of RAGFlow v0.23.0, bulk download is not supported, nor can you download an entire folder. 

docs/guides/manage_users_and_services.md

@@ -46,7 +46,7 @@ The Admin CLI and Admin Service form a client-server architectural suite for RAG
 2. Install ragflow-cli.
 
    ```bash
-   pip install ragflow-cli==0.22.1
+   pip install ragflow-cli==0.23.0
    ```
 
 3. Launch the CLI client:

docs/guides/models/deploy_local_llm.mdx

@@ -7,7 +7,7 @@ slug: /deploy_local_llm
 import Tabs from '@theme/Tabs';
 import TabItem from '@theme/TabItem';
 
-Deploy and run local models using Ollama, Xinference, or other frameworks.
+Deploy and run local models using Ollama, Xinference, VLLM ，SGLANG or other frameworks.
 
 ---
 
@@ -314,3 +314,41 @@ To enable IPEX-LLM accelerated Ollama in RAGFlow, you must also complete the con
 3. [Update System Model Settings](#6-update-system-model-settings)
 4. [Update Chat Configuration](#7-update-chat-configuration)
 
+### 5. Deploy VLLM 
+
+ubuntu 22.04/24.04
+
+```bash
+  pip install vllm 
+  ```
+### 5.1 RUN VLLM WITH BEST PRACTISE
+
+```bash
+nohup vllm serve /data/Qwen3-8B --served-model-name Qwen3-8B-FP8 --dtype auto --port 1025 --gpu-memory-utilization 0.90 --tool-call-parser hermes --enable-auto-tool-choice  > /var/log/vllm_startup1.log 2>&1 &
+  ```
+you can get log info
+```bash
+ tail -f -n 100 /var/log/vllm_startup1.log
+  ```
+when see the follow ,it means vllm engine is ready for access
+```bash
+Starting vLLM API server 0 on http://0.0.0.0:1025
+Started server process [19177]
+Application startup complete.
+ ```
+### 5.2 INTERGRATEING RAGFLOW WITH VLLM CHAT/EM/RERANK LLM WITH WEBUI
+
+setting->model providers->search->vllm->add ,configure as follow:
+
+![add vllm](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/ragflow_vllm.png)
+
+select vllm chat model as default llm model as follow:
+![chat](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/ragflow_vllm1.png)
+### 5.3 chat with vllm chat model
+create chat->create conversations-chat as follow:
+![chat](https://raw.githubusercontent.com/infiniflow/ragflow-docs/main/images/ragflow_vllm2.png)
+
+
+
+
+