Anthropic Proxy Rs

anthropic-proxy-rs

High-performance Rust proxy that translates Anthropic API requests to OpenAI-compatible format. Use Claude Code, Claude Desktop, or any Anthropic API client with OpenRouter, native OpenAI, or any OpenAI-compatible endpoint.

Features

- Fast & Lightweight: Written in Rust with async I/O (~3MB binary)
- Full Streaming: Server-Sent Events (SSE) with real-time responses
- Tool Calling: Complete support for function/tool calling
- Universal: Works with any OpenAI-compatible API (OpenRouter, OpenAI, Azure, local LLMs)
- Extended Thinking: Supports Claude's reasoning mode
- Drop-in Replacement: Compatible with official Anthropic SDKs

Quick Start
Note: Using Task is currently recommended. Install with brew install go-task (macOS) or see the installation guide. Releases with build binaries will be made soon.

Build and install to PATH with task

Install with curl

This installer uses cargo install --git ... --locked under the hood, so Rust/Cargo still needs to be present on the machine.

Run from anywhere

Or build and run manually

Configuration

Command Line Options

Commands:

Options:

Environment Variables

Configuration can be set via environment variables or .env file:

\* Required if your upstream endpoint needs authentication
\*\* The proxy automatically detects when a request has extended thinking enabled (via the thinking parameter in the request) and routes it to REASONING_MODEL. Standard requests without thinking use COMPLETION_MODEL. This allows you to use more powerful models for reasoning tasks and faster/cheaper models for simple completions. If not set, the model from the client request is used.

UPSTREAM_BASE_URL accepts any of these forms:
- Service base URL: https://api.openai.com -> /v1/chat/completions
- Versioned base URL:…