Polyclaude

Run Claude Code with any LLM provider — not just Anthropic.

Polyclaude is a transparent proxy that sits between Claude Code and your preferred model provider. It routes requests through LiteLLM so you can use GitHub Copilot, Google Gemini, Antigravity (Google's internal cloudcode API), or Anthropic models — all through Claude Code's familiar interface.

Why?

Claude Code normally only talks to Anthropic's API. If you have access to models through other providers (a Copilot subscription, a Gemini API key, an Antigravity-enabled Google account), you couldn't use them with Claude Code — until now.

Features

• Multi-provider support — GitHub Copilot, Google Gemini, Google Antigravity, and Anthropic
• Transparent to Claude Code — patches settings automatically so Claude Code talks to the local proxy
• Automatic model fallbacks — Claude Code's required model names (e.g. claude-sonnet-4) are silently mapped to an available provider when no Anthropic key is set
• OAuth authentication — GitHub device code flow, Google OAuth 2.0 with PKCE and auto-refresh, Anthropic API key entry
• Dynamic model discovery — fetches available models from each provider at runtime
• Antigravity interceptor — wraps/unwraps Google's cloudcode envelope format, with schema sanitization for cross-provider compatibility

Supported Providers

| Provider | Namespace | Auth | |---|---|---| | GitHub Copilot | copilot/<model> | Device code OAuth | | Google Gemini | gemini/<model> | Google OAuth 2.0 + PKCE | | Google Antigravity | antigravity/<model> | Google OAuth 2.0 + PKCE | | Anthropic | claude-* (native) | API key |

Installation

npm install -g polyclaude

Requires LiteLLM (pip install litellm) and Claude Code (npm install -g @anthropic-ai/claude-code).

Quick Start

1. Login to your provider(s)

polyclaude login

This walks you through authentication for each provider. You can set up one or all of them.

2. Run Claude Code with a specific model

# Use a Copilot model
polyclaude --model copilot/claude-sonnet-4

# Use a Gemini model
polyclaude --model gemini/gemini-2.5-pro

# Use an Antigravity model
polyclaude --model antigravity/gemini-3.1-pro-low
polyclaude --model antigravity/claude-opus-4-6-thinking

# Default — just run with whatever's available
polyclaude

Polyclaude starts the LiteLLM proxy in the background, patches Claude Code's settings to point at it, launches claude, and cleans up when you exit.

Commands

| Command | Description | |---|---| | polyclaude | Default — sync config, start proxy, launch Claude Code | | polyclaude setup | Sync model configs without launching Claude | | polyclaude login | Authenticate with providers | | polyclaude list | List all available models | | polyclaude start | Run the LiteLLM proxy in the foreground |

All other arguments are passed through to claude directly.

How It Works

Claude Code  →  LiteLLM proxy (port 4000)  →  Provider API
                     ↓
              Antigravity interceptor (port 51122)  →  cloudcode-pa.googleapis.com

1. Config sync — discovers available models from each authenticated provider and generates a LiteLLM config
2. Proxy startup — starts LiteLLM on port 4000 with the generated config, plus the Antigravity interceptor on port 51122
3. Settings patch — rewrites ~/.claude/settings.json to point ANTHROPIC_BASE_URL at the local proxy
4. Claude launch — starts claude CLI which now routes all API calls through the proxy
5. Cleanup — restores settings on exit

Available Antigravity Models

These models are available through Google's Antigravity/cloudcode API:

• gemini-3.1-pro-low, gemini-3.1-pro-high
• gemini-3-pro-low, gemini-3-pro-high
• gemini-3-flash, gemini-3.1-flash-image
• gemini-2.5-pro, gemini-2.5-flash
• claude-sonnet-4-6, claude-opus-4-6-thinking
• gpt-oss-120b-medium

Configuration

Config files are stored in ~/.polyclaude/ and ~/.litellm/:

• ~/.litellm/copilot-config.yaml — LiteLLM model routing config (auto-generated)
• ~/.litellm/.env — API keys and tokens
• ~/.claude/settings.json — patched by polyclaude to point at the local proxy

License

ISC