mirror of
https://github.com/open-webui/open-webui.git
synced 2026-05-06 10:58:17 -05:00
[GH-ISSUE #12656] issue: task model calls don't obey filter configuration #55339
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 @lattwood on GitHub (Apr 9, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/12656
Check Existing Issues
Installation Method
Docker
Open WebUI Version
v0.6.2
Ollama Version (if applicable)
No response
Operating System
N/A
Browser (if applicable)
No response
Confirmation
README.md.Expected Behavior
If I have a filter function added to a model, or a filter function enabled as global, I would expect that it is invoked for all uses of the model-
Actual Behavior
The filter function doesn't execute for task model invocations.
Steps to Reproduce
Logs & Screenshots
N/A
Additional Information
I think
open_webui.utils.filter.process_filter_functionsgets used for processing inlet and outlet filter functions, but inopen_webui/routers/tasks.pythe method is imported and never called.e31f680788This commit added the import of the
process_filter_functionsmethod, but the code modifications were all to add calls toprocess_pipeline_inlet_filter. This is speculation, but I think the intent of that commit was to add calls toprocess_filter_functions(filter_type="inlet", ...)?@lattwood commented on GitHub (Apr 10, 2025):
Also looks like this is the case with image generation
@lattwood commented on GitHub (Apr 11, 2025):
And this appears to be the case for when using the new MCPO tools- not when sending the initial request, but when o3-mini decides to start invoking tools during the reasoning process.
This is a problem because of using this inlet Filter to add user details to the outbound requests for cost tracking, which I've included below.
@csantv commented on GitHub (Apr 13, 2025):
i'm experiencing the same issue, switching to pipelines from filter functions could help but that is overkill as it'd require a new container
@tjbck commented on GitHub (Apr 14, 2025):
Intended behaviour, if we were to introduce this change, this will break a lot of existing filter workflows.