mirror of
https://github.com/open-webui/open-webui.git
synced 2026-07-22 11:44:11 -05:00
feat: persist tool calling & outputs in conversation memory #6174
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: github-starred/open-webui#6174
Reference in New Issue
Block a user
Originally created by @ralungei on GitHub (Aug 22, 2025).
Check Existing Issues
Problem Description
Open WebUI conversation memory only stores the LLM responses, but not the outputs from tools (e.g., web search results, API calls, calculator results). This causes the assistant to "forget" tool outputs, breaking the conversation flow and forcing redundant tool calls.
Desired Solution you'd like
Persist tool outputs in memory together with the LLM messages. When resuming or continuing a conversation, the assistant should have access to both the model text and tool responses, ensuring coherent and consistent interactions.
Alternatives Considered
No response
Additional Context
No response
@RodolfoCastanheira commented on GitHub (Aug 22, 2025):
This also breaks caching of most providers.
@jsweetzer-ea commented on GitHub (Aug 22, 2025):
For OpenAI models, the right solution is to support the Responses API.
@ralungei commented on GitHub (Aug 25, 2025):
For me breaks so many use cases or makes me ask the LLM to print internal ids
@johnnyasantoss commented on GitHub (Sep 25, 2025):
This still relevant and affects writing tools on open web ui as well.
@brandco commented on GitHub (Sep 25, 2025):
Could you use event_emitter to put tool call responses (or whatever you want) into the message content during the tool execution?
https://docs.openwebui.com/features/plugin/events/
@rgaricano commented on GitHub (Sep 25, 2025):
With Default Function Calling mode yes, with Native Function Calling mode no (no directly, but model can be prompted for add the results of the tool to message)