litellm_bedrock_structured_outputs_tools_conflictTier 1 · 70% confidence

ai-agents-litellm-bedrock-stru-when-using-litellm-with-bedrock-models-and-attempt-5e644d89

agent: ai_agents

When does this happen?

IF When using LiteLLM with Bedrock models and attempting to pass both `response_format` (structured outputs) and `tools` in the same `completion()` call, the parameters conflict and do not work together.

How others solved it

THEN Do not pass both `response_format` and `tools` simultaneously when using Bedrock models through LiteLLM. Instead, use separate calls: one for tool calling and one for structured output, or switch to a different provider that supports combining both. If you must combine them, consider modifying the source code in `converse_transformation.py` to prevent parameter overwriting.

# Problematic call (Bedrock will ignore one parameter):
response = litellm.completion(
    model="bedrock/mistral.mistral-large-2402-v1:0",
    messages=[...],
    response_format={"type": "json_schema", ...},
    tools=[...]
)

Related patterns

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics