Not Working with Google Gemini API

[devadattas]

I have the following config for mycoder.config.js:

export default {
    provider: 'openai',
    model: 'gemini-2.5-pro-exp-03-25',
    baseUrl: 'https://generativelanguage.googleapis.com/v1beta/openai',
    logLevel: 'verbose',
    githubMode: false,
    headless: true
};

I keep getting the following error:

Detected 2 browsers on the system
Available browsers:
- Chrome (chromium) at /Applications/Google Chrome.app/Contents/MacOS/Google Chrome
- Firefox (firefox) at /Applications/Firefox.app/Contents/MacOS/firefox
MyCoder v1.6.0 - AI-powered coding assistant
GitHub mode is enabled but there are issues with git/gh CLI tools:
- GitHub CLI is not available. Please install gh CLI.
GitHub mode requires git and gh CLI tools to be installed.
Please install the missing tools or disable GitHub mode with --githubMode false
Disabling GitHub mode due to missing CLI tools.
Looking API key in env: OPENAI_API_KEY
LLM: openai/gemini-2.5-pro-exp-03-25
Using API key: AIza...
Using base url: https://generativelanguage.googleapis.com/v1beta/openai

An error occurred:
Error: Error calling OpenAI API: 400 status code (no body)
Error: Error calling OpenAI API: 400 status code (no body)
    at OpenAIProvider.generateText (file:///Users/admin/.nvm/versions/node/v22.14.0/lib/node_modules/mycoder/node_modules/mycoder-agent/src/core/llm/providers/openai.ts:126:13)
    at processTicksAndRejections (node:internal/process/task_queues:105:5)
    at toolAgent (file:///Users/admin/.nvm/versions/node/v22.14.0/lib/node_modules/mycoder/node_modules/mycoder-agent/src/core/toolAgent/toolAgentCore.ts:142:45)
    at executePrompt (file:///Users/admin/.nvm/versions/node/v22.14.0/lib/node_modules/mycoder/src/commands/$default.ts:187:20)
    at Object.handler (file:///Users/admin/.nvm/versions/node/v22.14.0/lib/node_modules/mycoder/src/commands/$default.ts:280:5)
0
[Token Usage Total] Root: input: 0 cache-writes: 0 cache-reads: 0 output: 0 COST: $0.00
Forcing exit after 5000ms timeout

[devadattas]

@bhouston Please let me know if I am missing something while configuring?

[bhouston]

Issue Triage Analysis

Issue Classification

• Type: Bug
• Labels: bug, integration

Initial Assessment

The issue reports an error when trying to use mycoder with Google Gemini API via the OpenAI compatibility layer. The error occurs with the following configuration:

export default {
    provider: 'openai',
    model: 'gemini-2.5-pro-exp-03-25',
    baseUrl: 'https://generativelanguage.googleapis.com/v1beta/openai',
    logLevel: 'verbose',
    githubMode: false,
    headless: true
};

The error message is:

Error: Error calling OpenAI API: 400 status code (no body)

Duplication Check

This issue is related to and has spawned issue #391 "Fix Google Gemini OpenAI-compatible API integration" which proposes a solution.

Issue Investigation

Root Cause Analysis:

1. The OpenAI provider in mycoder is designed to work with OpenAI-compatible APIs, but there appear to be specific requirements for how authentication and API endpoints are structured for Google's Gemini API.

2. The current implementation in packages/agent/src/core/llm/providers/openai.ts initializes the OpenAI client with:

this.client = new OpenAI({
  apiKey: this.apiKey,
  ...(this.baseUrl && { baseURL: this.baseUrl }),
});

3. This approach works for some OpenAI-compatible APIs but appears to fail with Google's Gemini API, likely due to:
   • Different authentication requirements (Google might require an API key as a query parameter rather than in headers)
   • Potential differences in the request format or endpoint structure
   • Possible need for additional parameters specific to Google's API

Affected Components:

• OpenAI provider implementation (packages/agent/src/core/llm/providers/openai.ts)
• Provider configuration system

Possible Solutions:

1. Create a dedicated provider entry for Google Gemini in the provider registry
2. Modify the OpenAI provider to handle Google Gemini's specific authentication and endpoint requirements
3. Add more detailed error logging to help diagnose API compatibility issues

Next Steps

1. Review Google's Gemini API documentation for their OpenAI-compatible endpoint
2. Create a dedicated Gemini provider configuration that extends the OpenAI provider with Gemini-specific adjustments
3. Add more detailed error handling to capture and display the actual API request and response details

Recommendation

Based on the investigation, I recommend:

1. Adding a dedicated gemini provider entry similar to how xai is implemented
2. Modifying the OpenAI provider to better handle alternative authentication methods
3. Improving error logging to provide more detailed information about API failures

This appears to be a legitimate bug that requires code changes to support Google's Gemini API through their OpenAI-compatible endpoint.