qoder-proxy
Local OpenAI-compatible proxy for Qoder CN CLI, for learning only
A local proxy server that wraps the Qoder CLI, a Chinese AI coding service, and exposes it as a drop-in OpenAI or Anthropic-compatible API so any app that supports those formats can use Qoder models without code changes.
Qoder Proxy is a local adapter that wraps the Qoder CLI tool and exposes it as an HTTP server compatible with the OpenAI and Anthropic API formats. The Qoder CLI is a command-line interface for a Chinese AI coding service. This proxy lets local applications that know how to talk to OpenAI or Anthropic instead talk to Qoder, without any code changes on the client side.
The proxy only listens on the local machine (127.0.0.1) and is explicitly not intended for public deployment, shared use, or commercial resale. Users must supply their own Qoder account credentials. The readme opens with a disclaimer in both Chinese and English stressing that this is for personal protocol research only.
It supports two Qoder backends: a CN backend connecting to qoder.com.cn, authenticated with a personal access token, and a Global backend connecting to qoder.com, authenticated via OAuth login. You switch between them by changing a single environment variable. Supported models include Qwen3.7-Max, GLM-5.1, Kimi-K2.6, DeepSeek-V4 Pro, and several others.
The proxy exposes two API paths: one following the OpenAI chat completions format and one following the Anthropic messages format. It handles streaming output by forwarding the CLI stream as server-sent events. Tool calling is simulated by injecting format instructions into the prompt and parsing the text output, since the CLI does not have a native tool-calling channel. The readme notes this is less reliable than calling official APIs directly.
Installation requires Node.js 18 or later, plus installing the Qoder CLI package globally via npm. A small web console is available at a local URL for testing chat, viewing models, and checking local usage statistics. The project is released under the MIT license.
Where it fits
- Point any OpenAI-compatible desktop app at localhost to use Qoder AI models instead of the official OpenAI service.
- Use GLM-5.1 or Qwen3.7-Max through the Anthropic messages API format without changing any client code.
- Switch between Qoder's CN and Global backends by changing a single environment variable.
- Test streaming responses and tool-call simulations with Qoder models using the built-in local web console.