LunaProxy is a small program that runs on your own computer and pretends to be an OpenAI-style chat service, while actually talking to Qwen's web chat behind the scenes. The point is that any tool already set up to call OpenAI or Anthropic can be pointed at LunaProxy instead, and the requests get routed through to Qwen with no cloud relay in between. The project ships with a React admin dashboard that runs on the same port as the API, so configuration happens in a browser tab. The README is clear that this is for learning, research, and personal use. There is no commercial license, no warranty, and the author takes no responsibility for account suspensions, data loss, or any other consequence of running it. The feature list covers OpenAI-compatible endpoints at /v1/chat/completions and /v1/models, Anthropic-compatible endpoints at /v1/messages for Claude-style clients, server-sent-event streaming, and a tool-call bridge that wraps Qwen calls in an XML contract and translates the results back into OpenAI or Anthropic shapes. There is also queue and concurrency control across multiple Qwen accounts, persistent session and run tracking, and automatic offload of very large prompts to file-backed context. To run it you need Node.js 18 or newer, one of npm, pnpm, or Bun, and a set of Qwen credentials. Credentials can be captured automatically by letting LunaProxy open a real Chrome window for OAuth login, or pasted in manually as a token and cookie header. The default listening address is 127.0.0.1 on port 8080, and runtime settings live in data/config.json.
Generated 2026-05-21 · Model: sonnet-4-6 · Verify against the repo before relying on details.