Feature: Add Gemini model support

[weijiafu14]

Problem

This proxy currently only supports OpenAI-compatible backends. Users who want to use Gemini models (gemini-2.5-flash, gemini-3-pro, etc.) with Claude Code have no option here.

Given that Gemini offers free tier API access and competitive performance, Gemini support would be valuable.

Challenge

Gemini API is not OpenAI-compatible — it uses a completely different protocol (contents/parts instead of messages, functionCall/functionResponse instead of tool_calls, thoughtSignature for thinking mode, etc.). Simply pointing the OpenAI SDK at a Gemini endpoint won't work.

Possible approach

gemini-claude-bridge is an MIT-licensed library that handles the full Gemini ↔ Claude protocol conversion:

pip install gemini-claude-bridge

from gemini_claude_bridge import convert_claude_to_gemini, convert_gemini_to_claude, GeminiStreamConverter

It handles the tricky parts:

• thoughtSignature persistence for multi-turn tool calls
• Streaming tool arguments (partialArgs with jsonPath)
• Image handling in tool results
• JSON Schema cleaning for Gemini compatibility

Could be integrated as an optional backend alongside the existing OpenAI path, activated when the model name starts with gemini/.

Happy to help with a PR if there's interest.