Mixed Frontend and Backend Tool Execution Fails with OpenAI BadRequestError

[kittenkill]

Description

When executing both frontend and backend tools in a single user request, CopilotKit throws an OpenAI API error related to tool call ID validation. This issue affects both the deprecated coagents-starter example and the newer with-langgraph-python example.

The error occurs when a user requests actions that require both server-side execution (e.g., get_weather) and frontend UI updates (e.g., change_color) in the same prompt.

Examples Affected

• coagents-starter (original example)
• with-langgraph-python (modernized example)

Steps to Reproduce

1. Run this example
2. In the chat interface, enter a prompt that triggers both frontend and backend tools:

   Get the weather in SF, and change the color to brown
   

Expected Behavior

Both tools should execute successfully:

1. The backend get_weather tool should fetch weather data for San Francisco
2. The frontend change_color tool should update the UI color to brown
3. The assistant should provide a response summarizing both actions

Actual Behavior

The request fails with an OpenAI API error:

openai.BadRequestError: Error code: 400 - {
  'error': {
    'message': "Invalid parameter: 'tool_call_id' of 'call_qhvIZb4R9Jzv59TZPYBPnDih' not found in 'tool_calls' of previous message.", 
    'type': 'invalid_request_error', 
    'param': 'messages.[26].tool_call_id', 
    'code': None
  }
}

---

It seems that some message sequencing / ordering issue is happening when generating the requests to the LLM is triggering the issue.

May be related to CopilotKit/CopilotKit#561

This looks to be not a problem with the example per se, but with how its using copilotkit sdk or a copilotkit issue.

Thanks!

[julesale]

+1