Chat in the knowleadge base gives the source files, but the python code does not provide the source files

[Arun-Arcadix]

Hi Team,

This is my chat response from the knowledge base.

But I do the chat completion using the below code, I am not getting the source files.

try:
    from ragflow_sdk import RAGFlow  # Still need this to get chat_id
    rag_sdk_client = RAGFlow(api_key=RAGFLOW_API_KEY, base_url=RAGFLOW_BASE_URL)
    assistants = rag_sdk_client.list_chats(name=CHAT_ASSISTANT_NAME)
    if not assistants:
        print(f"No assistant found with name: {CHAT_ASSISTANT_NAME}. Please create one in RAGFlow UI.")
        exit()
    assistant = assistants[0]
    chat_id = assistant.id  # Get the ID of your chat assistant
    print(f"Found chat assistant: {assistant.name} with ID: {chat_id}")

    # Now, initialize the OpenAI client with the correct base_url
    openai_client = OpenAI(
        api_key=RAGFLOW_API_KEY,  # RAGFlow usually uses this for auth
        base_url=f"{RAGFLOW_BASE_URL}/api/v1/chats_openai/{chat_id}"  # Specific OpenAI-compatible endpoint
    )
    print("OpenAI client pointed to RAGFlow initialized successfully.")
    chat_history= []
    chat_history.append({"role": "user", "content": user_message})
    completion = openai_client.chat.completions.create(
        model="model",  # This is often a placeholder for RAGFlow's OpenAI API.
        # The actual model is determined by the RAGFlow assistant config.
        messages=chat_history,
        stream=False,
        extra_body={"show_ref_docs": True}
    )
    # ⭐️ NEW: Get the raw dictionary representation of the completion object
    raw_completion_data = completion.model_dump()

    # RAGFlow often places ref_docs here:
    references_data = raw_completion_data.get('extra', {}).get("ref_docs", [])

    assistant_response = completion.choices[0].message.content

    chat_history.append({"role": "assistant", "content": assistant_response})
    print(assistant_response)
    return assistant_response
except Exception as e:
    print(f"Error initializing RAGFlow or OpenAI client: {e}")

This is my raw_completion_data, Please help.

{'choices': [{'finish_reason': 'stop', 'index': 0, 'logprobs': None, 'message': {'annotations': None, 'audio': None, 'content': 'The knowledge base contains information about the PopupFrame component, which is a React component used in the Arcadix framework. Here's a summary of its functionality and structure:

ID: 0

• Purpose: The PopupFrame component serves as a container or frame for displaying popup content within the Arcadix application. It handles the positioning, sizing, and basic UI elements (like maximize and close buttons) for the popup.
• Imports:
  • React: For creating the component.
  • Base_Hook: Provides base class functionality and a reducer for managing the component's state.
  • PopupFrame_Hook: Contains logic for initializing the component's state.
  • PopupComponentController: Used to dynamically load the content to be displayed within the popup.
  • Popup_TextResource: For managing the text displayed in the popup (e.g., title, subtitle).
  • Images: Imports inline images for the maximize and close icons.
• State Management:
  • Uses the useReducer hook to manage the component's state, including properties like height, width, position (left, top), and maximization status.
  • The initial state is determined by PopupFrame_Hook.GetInitialState(props).
• Dragging Functionality:
  • Allows the popup to be dragged around the screen.
  • Uses OnMouseDown, OnMouseMove, and OnMouseUp event handlers to manage the dragging behavior.
  • The isClicked attribute on the frame is used to determine if a drag is in progress.
• Resizing:
  • Implements window resize event listener to adjust popup position.
• Maximizing/Minimizing:
  • Provides a ToggleMaximize function to switch the popup between a maximized (100% height and width) and a normal size.
• Dynamic Content Loading:
  • Uses PopupComponentController to load the actual content of the popup dynamically.
  • The useLayoutEffect hook is used to load the popup component and get its dimensions.

ID: 1

• GetContent function:
  • Renders the Popup's JSX structure, which includes:
    • A modal backdrop (div className="modal")
    • The popup frame itself (div className="popup-innner-modal")
    • A header (div className="popup-header") with:
      • Title and subtitle (if props.Data.PopupData.Meta.ShowHeader is true)
      • Maximize and close icons (conditionally rendered based on props)
    • The actual popup content, rendered using the PopupComponent component.
• Event Handling:
  • The header includes onMouseDown, onMouseUp, onTouchStart, and onTouchEnd handlers for dragging.
  • The close icon has an onClick handler that calls the ClosePopup event.
• PopupComponent:
  • Renders the content of the popup, passing in relevant data, callbacks, and events.
• React.memo:
  • The component is wrapped in React.memo to prevent unnecessary re-renders. It checks if the Id prop has changed to determine if a re-render is needed.
• Z-Index:
  • If there is no zindex defined then it defaults to 1003.

In summary, PopupFrame is a reusable component that provides the basic structure and functionality for displaying dynamic popup content in the Arcadix framework. It handles positioning, dragging, maximizing, closing, and loading the actual popup content.', 'function_call': None, 'refusal': None, 'role': 'assistant', 'tool_calls': None}}], 'created': 1764928521, 'id': 'chatcmpl-72c6797ac91c11f0a6fe261421afc4ce', 'model': 'model', 'object': 'chat.completion', 'service_tier': None, 'system_fingerprint': None, 'usage': {'completion_tokens': 3386, 'completion_tokens_details': {'accepted_prediction_tokens': 3386, 'audio_tokens': None, 'reasoning_tokens': 24, 'rejected_prediction_tokens': 0}, 'prompt_tokens': 24, 'prompt_tokens_details': None, 'total_tokens': 3410}}

[dosubot[bot]]

Hi @Arun-Arcadix! I'm Dosu and I’m helping the ragflow team.

The reason you aren't seeing source file references in your Python code is that the OpenAI-compatible endpoint expects the parameter to be named reference (not show_ref_docs) to include source references in the response. Update your code to use extra_body={"reference": True} in your chat completion request. When enabled, the references will appear in choices[0].message.reference in the response for non-streaming mode (source).

If you still don't see references after this change, there is a known issue where the backend or SDK may omit them. Fixes involve backend and SDK updates to ensure the reference field is included in the response (details and code pointers). After updating, restart your service and make sure the quote parameter is set to true in your assistant config.

Once working, you can extract file paths or document names from the reference field in the response. If your knowledge base indexing includes file paths as metadata, they will be present in the reference data.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[Arcadix]

@dosu what steps we need to follow on Backend or SDK Updates, Can you Please Guide us?
or any Documentation?

[dosubot[bot]]

To enable source references in your RAGFlow Python SDK responses, you need to:

1. Update your backend to at least v0.20.4, as this version includes fixes for reference metadata in OpenAI-compatible completions and file preview/citation wiring. If you’re running an older version, upgrade and redeploy your backend service (API docs).

2. In your Python code, use extra_body={"reference": True} (not show_ref_docs) when calling chat.completions.create. The references will appear in choices[0].message.reference for non-streaming responses (Python SDK docs).

3. Make sure your chat assistant config has show_quote set to true. This ensures the backend includes source metadata in responses (API docs).

4. If you still don’t see references, you may need to patch your backend code:

   • In ragflow/api/apps/sdk/session.py, before yielding the response in the chat completions endpoint, add:

     reference = ans.get("reference")
     response["reference"] = reference

   • In ragflow/api/db/services/dialog_service.py, ensure that when streaming, the yielded dictionary includes the reference field with source document metadata (example).

5. After making code changes, rebuild and restart your backend (for Docker: docker compose -f docker/docker-compose.yml up -d ragflow).

6. When working, the reference field in the response will contain document_name, document_id, and other metadata for each source chunk.

If you want to see full code examples or more details, check the Python API reference and related GitHub issue.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[Magicbook1108]

try what dosu said. If you still cannot view the reference, then you might have to raise a feature request for this issue.