wholock2210 / Luna-Proxy

What is Luna-Proxy?

Luna-Proxy is a local reverse proxy that exposes Alibaba's Qwen web chat as an OpenAI-compatible API. Built in JavaScript with a Koa backend and React admin interface, it lets you route requests through Qwen from any OpenAI-compatible client--think Claude Code, Cline, or any custom tool--without touching Qwen's native API directly. The proxy handles credential management, multi-account concurrency, session persistence, and even prompt overflow by offloading large contexts to file-backed storage. Credentials are captured either through automated browser OAuth or manual token/cookie input, with a React dashboard for configuration and monitoring.

Why is it gaining traction?

The hook is simple: you get free access to Qwen models through a familiar API contract. Developers tired of hunting for API keys or managing provider accounts can point their existing tooling at localhost and go. The built-in admin UI is unusually complete for a proxy project--providers, sessions, runs, network profiles, and logs are all inspectable in one place. Tool-call bridging is the other differentiator: the proxy injects an XML contract into prompts and parses Qwen's tool outputs back into OpenAI or Anthropic formats, so clients expecting function calling work without modification.

Who should use this?

This is for developers using OpenAI-compatible clients who want to experiment with Qwen without leaving their existing workflow. Agents and coding assistants expecting a standard `/v1/chat/completions` endpoint will work out of the box. Teams running multiple accounts or needing IP isolation for provider routing will find the scheduler and worker management useful. If you need a local proxy that acts like a drop-in gateway with observability baked in, this fits.

Verdict

At 10 stars with a credibility score of 0.699999988079071%, Luna-Proxy is early-stage and unproven at scale. The feature surface is broad, the documentation is clear, and the React UI suggests thoughtful UX, but test coverage and community feedback are minimal. Worth trying for local development or experimentation, but do not deploy to production without thorough validation.