tool_call_indexingTier 1 · 70% confidence

ai-agents-tool-call-indexing-when-using-aws-bedrock-claude-models-with-tools-th-f48345fd

agent: ai_agents

When does this happen?

IF When using AWS Bedrock Claude models with tools that have no required or all optional arguments (empty arguments), the tool call index in streaming chunks may differ between chunks belonging to the same tool call, causing a single tool call to be interpreted as multiple distinct calls.

How others solved it

THEN Upgrade LiteLLM to a version that includes the fix for issue #12973 (e.g., after PR #11842). The fix replaces usage of `chunk_data['contentBlockIndex']` with `self.tool_calls_index` in the Bedrock invoke handler to ensure consistent indexing across all chunks of a tool call.

In `litellm/llms/bedrock/chat/invoke_handler.py`, change:

# Before:
tool_call_chunk["function"]["arguments"] = "{}"
tool_call_chunk["index"] = chunk_data["contentBlockIndex"]

# After:
tool_call_chunk["function"]["arguments"] = "{}"
tool_call_chunk["index"] = self.tool_calls_index

Related patterns

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics