It crashes

[sayanb]

I just came across this project and wanted to give it a go. Apologies if this is not the right way to run this! I am running this under the root directory of a Python FastAPI project

1. export OPENAI_API_KEY=<my-key>
2. mycoder --provider openai --model gpt-4o -i
3. > explain what endpoint <one of the api endpoints> does

An error occurred:
Error: Error calling OpenAI API: 400 An assistant message with 'tool_calls' must be followed by tool messages responding to each 'tool_call_id'. The following tool_call_ids did not have response messages: <tool-call-id>
Error: Error calling OpenAI API: 400 An assistant message with 'tool_calls' must be followed by tool messages responding to each 'tool_call_id'. The following tool_call_ids did not have response messages: <tool-call-id>
    at OpenAIProvider.generateText

[bhouston]

Issue Analysis

Thank you for reporting this issue! I've investigated the crash and here's what I found:

Root Cause

The error occurs because the OpenAI API requires that every tool call from an assistant message must be followed by a tool response message with the corresponding tool_call_id. In your case, when using MyCoder with the gpt-4o model in interactive mode, the system is likely sending an incomplete conversation history to the OpenAI API where a previous tool call doesn't have a corresponding tool response.

Technical Details

The issue is in how MyCoder formats messages for the OpenAI API. When you're in interactive mode and send a new prompt, the conversation history might include tool calls that haven't been properly responded to, which violates OpenAI's API requirements.

The specific error message:

Error calling OpenAI API: 400 An assistant message with 'tool_calls' must be followed by tool messages responding to each 'tool_call_id'. The following tool_call_ids did not have response messages: <tool-call-id>

indicates that the API is rejecting the request because it contains a tool call without a corresponding response.

Possible Solutions

1. Modify the formatMessages method in the OpenAIProvider class to ensure every tool call has a corresponding tool response before sending to the API
2. Add validation in the interactive mode to check the conversation state
3. Implement error handling specifically for this type of OpenAI API error

Workaround

Until this is fixed, you might be able to work around the issue by:

• Using a different model that's less likely to use tool calls (e.g., gpt-3.5-turbo)
• Using a different provider (e.g., Claude)
• Avoiding interactive mode with the OpenAI provider

I'll add appropriate labels to this issue and prioritize it for fixing.