OpenProxy

中文 | English

[!NOTE]
Complete rewrite with semantic IR architecture, supporting multi-provider bidirectional conversion.

OpenProxy is an API translation gateway that converts between different AI providers (Anthropic, Gemini, OpenAI). Use any client to call any provider.

Features

• Multi-Provider Support: Anthropic, OpenAI (Chat/Responses), Google Gemini
• Bidirectional Conversion: Use Claude client to call GPT, or OpenAI client to call Claude
• Semantic IR: Internal abstraction layer for clean protocol conversion
• Streaming Support: Real-time streaming for all providers
• Tool Calling: Function calling across providers
• Vision Support: Image input for compatible models
• Built-in Tools: Support for web_search, file_search, etc.
• Zero Config: Client-driven configuration via Authorization header
• Official SDKs: Uses official provider SDKs for reliability

Quick Start

Installation

# Clone repository
git clone https://github.com/yourusername/openproxy.git
cd openproxy

# Install dependencies
npm install

# Run in development mode
npm run dev

# Or build and run
npm run build
npm start

Service runs at http://localhost:3000

Web Config Generator

Open http://localhost:3000 in your browser to use the interactive configuration generator:

1. Select your target provider (OpenAI, Anthropic, Gemini)
2. Enter your API credentials
3. Copy the generated configuration for your CLI tool

Basic Usage

Using Claude CLI with OpenAI

# Set up environment
export ANTHROPIC_BASE_URL="http://localhost:3000"
export ANTHROPIC_AUTH_TOKEN='{"key":"sk-your-openai-key","url":"https://api.openai.com/v1","target":"openai-chat"}'
export ANTHROPIC_MODEL="gpt-4o"

# Use Claude CLI
claude

Using Codex CLI with OpenAI Responses API

export OPENAI_API_KEY='{"key":"sk-your-openai-key","url":"https://api.openai.com/v1","model":"gpt-4o","target":"openai-responses"}'
codex --yolo -c model_providers.openproxy.base_url="http://localhost:3000" \
      -c model_provider="openproxy" \
      -c model="gpt-4o"

Using Gemini CLI with Anthropic

export GOOGLE_GEMINI_BASE_URL="http://localhost:3000"
export GEMINI_API_KEY='{"key":"sk-ant-your-key","url":"https://api.anthropic.com/v1","model":"claude-3-5-sonnet-20241022","target":"anthropic"}'
gemini -y "Hello!"

Documentation

• 📚 Docs Index - Central documentation entry
• 👤 User Guide - Quick start and examples
• 🧑‍💻 Developer Guide - Development workflow
• 🏗️ Architecture - Technical architecture
• 🧪 Testing Guide - Testing overview and commands

Supported Matrix

Clients → Targets

Features

Configuration Format

API Keys are provided by clients via HTTP Headers as JSON:

{
  "key": "your-actual-api-key",
  "url": "https://api.openai.com/v1",
  "model": "gpt-4o",
  "target": "openai-chat"
}

Pass this JSON as:

• Authorization: Bearer <json> for OpenAI/Anthropic clients
• x-goog-api-key: <json> for Gemini clients

URL Format

http://localhost:3000/<source-api-path>

Examples:

• /v1/messages - Anthropic request format
• /v1/chat/completions - OpenAI Chat request format
• /v1beta/models/gemini-pro:generateContent - Gemini request format

Target provider is specified via the target field in the header config.

Testing

# L0-L2
npm test

# L3 SDK tests (auto-starts server)
npm run test:3

# L4 CLI tests (auto-starts server; requires CLIs installed)
npm run test:4

# All (L0-L4)
npm run test:all

# Selective run
npx tsx src/test/test-runner.ts --level=4 --cli=claude --target=anthropic
L4_TASK=file-write npx tsx src/test/test-runner.ts --level=4

Project Structure

src/
  core/              # IR types and proxy logic
    ir.ts            # Semantic IR definitions
    proxy.ts         # Main proxy logic
    provider.ts      # Provider interface
  providers/         # Provider implementations
    anthropic/       # Anthropic provider
    gemini/          # Gemini provider
    openai-chat/     # OpenAI Chat Completions
    openai-responses/# OpenAI Responses API
  routing/           # URL routing and config parsing
    router-v2.ts     # Request router
    config.ts        # Configuration parsing
  test/              # Test suites
    0-provider/      # L0: Provider health checks
    1-converter/     # L1: Format converters
    2-api/           # L2: API integration tests
    3-lib/           # L3: SDK tests
    4-cli/           # L4: CLI end-to-end tests

Docker

docker build -t openproxy .
docker run -p 3000:3000 openproxy

License

CC-BY-NC-4.0