Bug Description
Admin Settings -> Interface configurations are intended to be global settings that apply to ALL models in the instance.
However, when I disable these parameters (Auto-tags, Title Generation, Suggestions) in the Admin panel, the restriction is not applied to the specific model I selected (my Manifold pipeline function).
Behavior
When features are disabled in Admin Settings -> Interface, they should be disabled globally for every model,
including Manifold pipelines and external functions. The selected model should never receive
Task: prompts for these features when the global switch is False.
Expected Behavior:
When features are disabled, this will apply to the specific model I selected.
The disable features:
Follow-up suggestions
Chat title generation
Tag generation
This causes a single user message to trigger 4 separate executions of the pipeline/function, which is problematic for agentic pipelines or metered APIs.
Steps to Reproduce
Create a function/pipeline using the manifold type (e.g., a simple agent wrapper).
Disable the following in Admin Settings -> Interface:
Follow-up suggestions
Chat title generation
Tag generation
Select the specific model served by your pipeline in the chat interface.
Send a single user message (e.g., "Hello").
Observe the logs of the pipeline/function(The pipeline should execute once for the user message).
Select another model in the chat interface.
Send a single user message (e.g., "Hello").
All the features that I disabled applied: the Follow-up, Chat title and Tag were gone.
Behavior:
The pipeline should execute once for the user message in the specific model.
When I enable that features:
Actual Behavior
The pipeline receives 4 sequential requests:
The user message
Task: Suggest 3-5 relevant follow-up questions...
Task: Generate a concise, 3-5 word title...
Task: Generate 1-3 broad tags...
Logs / Screenshots
(From pipeline logs)
[...T12:06:45] === PIPE EXECUTION 1 START === (User message: "Proj-32786")[...T12:06:51] === PIPE EXECUTION 2 START === (User message: "Task: Suggest 3-5 relevant follow-up questions...")[...T12:06:51] === PIPE EXECUTION 3 START === (User message: "Task: Generate a concise, 3-5 word title...")[...T12:06:51] === PIPE EXECUTION 4 START === (User message: "Task: Generate 1-3 broad tags...")
Environment
Open WebUI Version: 0.7.2
Browser: Chrome
Related to Pipelines Issue #386 . The current workaround requires implementing specific logic inside the pipeline pipe() method to detect and reject system prompts containing.
Task: but this should be handled at the application level to prevent unnecessary overhead.
Bug Description
Admin Settings -> Interface configurations are intended to be global settings that apply to ALL models in the instance.
However, when I disable these parameters (Auto-tags, Title Generation, Suggestions) in the Admin panel, the restriction is not applied to the specific model I selected (my Manifold pipeline function).
Behavior
When features are disabled in Admin Settings -> Interface, they should be disabled globally for every model,
including Manifold pipelines and external functions. The selected model should never receive
Task: prompts for these features when the global switch is False.
Expected Behavior:
When features are disabled, this will apply to the specific model I selected.
The disable features:
Follow-up suggestions
Chat title generation
Tag generation
This causes a single user message to trigger 4 separate executions of the pipeline/function, which is problematic for agentic pipelines or metered APIs.
Steps to Reproduce
Create a function/pipeline using the manifold type (e.g., a simple agent wrapper).
Disable the following in Admin Settings -> Interface:
Follow-up suggestions
Chat title generation
Tag generation
Select the specific model served by your pipeline in the chat interface.
Send a single user message (e.g., "Hello").
Observe the logs of the pipeline/function(The pipeline should execute once for the user message).
Select another model in the chat interface.
Send a single user message (e.g., "Hello").
All the features that I disabled applied: the Follow-up, Chat title and Tag were gone.
Behavior:
The pipeline should execute once for the user message in the specific model.
When I enable that features:
Actual Behavior
The pipeline receives 4 sequential requests:
The user message
Task: Suggest 3-5 relevant follow-up questions...
Task: Generate a concise, 3-5 word title...
Task: Generate 1-3 broad tags...
Logs / Screenshots
(From pipeline logs)
[...T12:06:45] === PIPE EXECUTION 1 START === (User message: "Proj-32786")[...T12:06:51] === PIPE EXECUTION 2 START === (User message: "Task: Suggest 3-5 relevant follow-up questions...")[...T12:06:51] === PIPE EXECUTION 3 START === (User message: "Task: Generate a concise, 3-5 word title...")[...T12:06:51] === PIPE EXECUTION 4 START === (User message: "Task: Generate 1-3 broad tags...")
Environment
Open WebUI Version: 0.7.2
Browser: Chrome
Related to Pipelines Issue #386 . The current workaround requires implementing specific logic inside the pipeline pipe() method to detect and reject system prompts containing.
Task: but this should be handled at the application level to prevent unnecessary overhead.