NVIDIA Anthropic Proxy

中文文档

A Cloudflare Worker proxy that enables Claude Code to use NVIDIA NIM API models.

• Use NVIDIA NIM open-source models (Llama, Minimax, GLM, etc.) in Claude Code
• Seamless model switching with full Claude Code experience
• Low-latency access via Cloudflare's global edge network

Quick Start

1. Deploy the Proxy

git clone https://github.com/evanlong-me/nvidia-anthropic-proxy.git
cd nvidia-anthropic-proxy
npm run setup
npm run deploy

Prerequisites:

• NVIDIA API Key - Required during npm run setup
• Cloudflare account - npm run setup will prompt to login on first run (or set CLOUDFLARE_API_TOKEN env var for non-interactive use)

After deployment, you'll get a Worker URL: https://nvidia-anthropic-proxy.xxx.workers.dev

2. Configure Claude Code

Edit ~/.claude/settings.json:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://nvidia-anthropic-proxy.xxx.workers.dev",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "minimaxai/minimax-m2.1",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "minimaxai/minimax-m2.1",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "z-ai/glm4.7"
  }
}

3. Start Using

claude

• /model opus → minimaxai/minimax-m2.1
• /model sonnet → minimaxai/minimax-m2.1
• /model haiku → z-ai/glm4.7

Supported Models

The proxy uses pass-through mode, supporting all NVIDIA NIM models:

Full list: build.nvidia.com/models

How It Works

Claude Code (Anthropic format)
    ↓
This Proxy (Cloudflare Worker)
    ↓
Format Conversion (Anthropic → OpenAI)
    ↓
NVIDIA NIM API

Local Development

npm run dev   # Start local server
npm run tail  # View real-time logs

Star History

License

MIT