Package Exports

codex-proxy
codex-proxy/src/anthropic-proxy.js

This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (codex-proxy) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

Readme

Codex Proxy - Claude Code to GLM-4.5-Air

A format conversion proxy server that enables using GLM-4.5-Air model with Claude Code CLI including full tool calling support.

What It Does

This proxy acts as a bridge between Claude Code and the GLM-4.5-Air model with format conversion:

✅ Receives requests from Claude Code in Anthropic's Messages API format
✅ Converts to OpenAI Chat Completions format
✅ Forwards to GLM-4.5-Air API
✅ Converts responses back to Anthropic format
✅ Full tool calling support (Anthropic tool_use ↔ OpenAI tool_calls)
✅ Handles both streaming and non-streaming responses
✅ Comprehensive logging for debugging

Result: Use Claude Code with FREE GLM-4.5-Air model with full functionality!

Quick Start (NPM Installation)

3 simple steps to get started:

1. Install globally via npm

npm install -g codex-proxy

2. Start the proxy

codex-proxy              # Normal mode
codex-proxy --debug      # Debug mode (verbose logging)
codex-proxy --port 4000  # Custom port
codex-proxy --help       # Show all options

3. Configure Claude Code

In a new terminal:

export ANTHROPIC_AUTH_TOKEN="dummy"
export ANTHROPIC_BASE_URL="http://localhost:3333"
claude

That's it! Start using Claude Code with free Chutes GLM models.

Local Development Setup

Install dependencies
```
npm install
```

Configure environment variables

cp .env.example .env
# Edit .env with your GLM-4.5-Air API credentials

Start the proxy server

Normal mode (minimal logging):
```
npm run proxy
```
Debug mode (verbose logging):
```
npm run proxy:debug
```

Configure Claude Code

To use the proxy with Claude Code CLI:

Option 1: Using environment variables (Recommended)

export ANTHROPIC_AUTH_TOKEN="dummy"
export ANTHROPIC_BASE_URL="http://localhost:3333"
claude

Option 2: Update ~/.claude/settings.json

{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "dummy",
    "ANTHROPIC_BASE_URL": "http://localhost:3333"
  }
}

Usage

Start the proxy server (in one terminal):

npm run proxy
# or for debug mode with full logs:
npm run proxy:debug

Use Claude Code normally (in another terminal):
```
export ANTHROPIC_AUTH_TOKEN="dummy"
export ANTHROPIC_BASE_URL="http://localhost:3333"
claude
```
Claude Code will now use GLM-4.5-Air model with full tool calling support!

To switch back to regular Claude:

unset ANTHROPIC_AUTH_TOKEN ANTHROPIC_BASE_URL
claude

Configuration

The proxy uses the following environment variables (in .env):

GLM_API_TOKEN: Your GLM-4.5-Air API token (required)
GLM_API_URL: GLM-4.5-Air API endpoint (default: https://llm.chutes.ai/v1/chat/completions)
GLM_MODEL: Model to use (default: zai-org/GLM-4.5-Air)
PORT: Proxy server port (default: 3333)

Features

✅ Full tool calling support - Anthropic tool_use ↔ OpenAI tool_calls conversion ✅ Streaming responses - Real-time token-by-token output ✅ Comprehensive logging - Color-coded debug logs for troubleshooting ✅ Format conversion - Seamless Anthropic ↔ OpenAI translation ✅ Free GLM-4.5-Air API - Use GLM-4.5-Air at no cost ✅ Multiple models - Maps to Haiku, Sonnet, and Opus tiers

Logging

The proxy provides color-coded logs for easy debugging:

📥 Blue - Incoming requests from Claude Code
🔄 Magenta - Format conversion operations
📤 Cyan - Requests forwarded to Chutes API
✅ Green - Successful operations
❌ Red - Errors and failures
🌊 Cyan - Streaming responses

Normal mode: Clean, minimal logs Debug mode: Full request/response bodies and conversion details

Documentation

RESEARCH.md - Complete research findings and solution approaches
TEST-RESULTS.md - GLM-4.5-Air API test results
TESTING-GUIDE.md - Comprehensive guide for testing and debugging the proxy

Support

For issues or inquiries, contact: saradhi8142385201@gmail.com

License

MIT