mirror of
https://github.com/open-webui/open-webui.git
synced 2026-05-07 11:28:35 -05:00
[GH-ISSUE #21178] bug: Pipes: Broken RAG Access & Inconsistent Task Payloads for Custom Agents Pipe for GPT 5.2 #58077
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: github-starred/open-webui#58077
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @Soliver84 on GitHub (Feb 5, 2026).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/21178
Check Existing Issues
Installation Method
Pip Install
Open WebUI Version
v.0.7.2
Ollama Version (if applicable)
No response
Operating System
Windows Server 2022
Browser (if applicable)
No response
Confirmation
README.md.Expected Behavior
When using a Pipe Function (Manifold) to integrate a custom model (e.g., an Agent acting as "GPT-5.2"):
RAG Access: The model should be able to execute the system tool query_knowledge_files (passed via tools) to retrieve chunks from the active Knowledge Base, just like native models do.
Task Handling: Background tasks (Title Generation, Tag Generation, Follow-up Questions) should send a consistent payload to the Pipe, specifically including the input field as required by the OpenAI Responses API schema (and OpenWebUI's own _run_task_model_request helper).
Model IDs: When OpenWebUI calls the Pipe model (e.g., for a task), it should strip the internal manifold prefix or the Pipe framework should handle the prefixed ID automatically, so external API calls don't fail with model_not_found.
Actual Behavior
RAG Failure: The Pipe receives the tool definition for query_knowledge_files, but when the model attempts to call it (via tools or event_call), the execution fails or returns no data. The agent effectively has no access to the uploaded documents.
Task Crashes: Background tasks crash with ValueError: Task request missing 'input' content for Responses API. The payload sent to the Pipe contains messages but lacks the input field required by the internal helper function.
Model ID Error: During tasks, OpenWebUI sends the full prefixed ID (e.g., openai_responses_manifold_main_suche.gpt-5.2-thinking-high). Passing this ID to the OpenAI API results in 400 Bad Request: model_not_found.
Steps to Reproduce
Setup: Create a new Function (Pipe) in OpenWebUI using the openai_responses template or a custom Agent script.
Configuration: Configure the Pipe to act as a Manifold (Model), e.g., "GPT-5.2".
RAG Test:
Upload a document to the Knowledge Base.
Activate the Knowledge Base in a chat with the Pipe Model.
Ask a question about the document.
Result: The model cannot read the document (Tool execution fails or returns empty).
Task Test:
Send a message in the chat.
Wait for Title Generation or click "Generate Tags".
Result: Check the Logs. The task fails with ValueError: Task request missing 'input'.
Result 2: If input is patched manually, the API call fails with model_not_found because the model ID includes the Manifold prefix.
Logs & Screenshots
File "/app/backend/open_webui/routers/tasks.py", line 312, in generate_follow_ups
return await generate_chat_completion(request, form_data=payload, user=user)
...
ValueError: Task request missing 'input' content for Responses API.
aiohttp.client_exceptions.ClientResponseError: 400, message='Bad Request', url='https://api.openai.com/v1/responses'
RuntimeError: 400 Bad Request: {"error": {"message": "The requested model 'openai_responses_manifold_main_suche.gpt-5.2-thinking-high' does not exist.", "type": "invalid_request_error", "param": "model", "code": "model_not_found"}}
Additional Information
Additional Information
I have developed a workaround script (V19) that manually patches these issues:
It strips the model prefix.
It copies messages to input for tasks.
It implements a complex "RAG Pass-Through" logic to handle query_knowledge_files.
However, these should be handled natively by the OpenWebUI Pipe/Manifold architecture to allow developers to build robust Agents without hacking the pipe() method extensively.
@pr-validator-bot commented on GitHub (Feb 5, 2026):
⚠️ Missing Issue Title Prefix
@Soliver84, your issue title is missing a prefix (e.g.,
bug:,feat:,docs:).Please update your issue title to include one of the following prefixes:
Example:
bug: Login fails when using special characters in password@Soliver84 commented on GitHub (Feb 5, 2026):
nix!
@Soliver84 commented on GitHub (Feb 5, 2026):
Pipecode.txt