Skip to content

Conversation

Thomvis
Copy link

@Thomvis Thomvis commented Jul 25, 2025

This PR adds support for passing tools through the Configuration. I built this because I wanted to run Open Deep Research from within my own Python application and give it my own tools, without the overhead of MCP.

@rlancemartin rlancemartin force-pushed the main branch 2 times, most recently from 214871e to c84ad97 Compare August 8, 2025 05:28
@rlancemartin
Copy link
Collaborator

@Thomvis thank you! we recently merged some updates. you can please resolve conflicts? we also may want the ability to update the research system prompt here with configurable instructions re: the new tools.

@Thomvis Thomvis force-pushed the feature/custom_tools branch from 4f79b3c to 10e2e80 Compare September 5, 2025 10:16
@Thomvis
Copy link
Author

Thomvis commented Sep 5, 2025

I've rebased my work on top of main and added a custom_tools_prompt property to Configuration that can be used to add info about the custom tools to the research system prompt.

Testing revealed that Gemini 2.5 Pro intermittently returns
MALFORMED_FUNCTION_CALL finish reasons when handling complex tool-use
scenarios with multiple tools and detailed system prompts.
The presence of a meta-cognition tool (think_tool) alongside
action-oriented tools appears to create ambiguity in the model's
response generation.
Since the API doesn't return the malformed function call itself for debugging,
we've simplified the tool set by removing think_tool,
which should reduce the model's confusion and improve reliability.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants