Unable to Pass Custom Parameters to LLM provider Models

[kuyu12]

I am unable to find a way to pass custom parameters into a model when using DSPy.

For example, with AWS Bedrock, I want to pass a performanceConfig parameter but couldn’t find a supported way to do so. With boto3, the equivalent works as follows:

client.invoke_model( modelId=modelId, body=body, performanceConfigLatency="optimized" )

Similarly, I want to pass tool configurations (e.g., web search) for Claude or OpenAI models. For example, the Claude API allows:
tools=[{ "type": "web_search_20250305", "name": "web_search", "max_uses": 5 }]

I tried several different approaches in DSPy, but none worked.

Question:
Is it currently possible to pass such parameters or tool configurations through DSPy? If yes, how should it be implemented?

Environment:
DSPy version: 3.0.0b2 or 2.6.27
Python version: 3.12
Model provider: AWS Bedrock / Claude / OpenAI

[haddyadnan]

Have you tried this https://dspy.ai/api/modules/ReAct/?

[kuyu12]

ReAct is kind of give solution for the tools,
But I still have issue with the custom param, for exmple "performanceConfigLatency" I didnt find any way to pass it.

[rayanehmi]

Would love to be able to use provider web search --> https://docs.litellm.ai/docs/completion/web_search