oy-paddy/aliyun-model-proxy

This project is a reverse proxy server for AI language models hosted by Alibaba Cloud's DashScope service. A reverse proxy sits between your software and the actual AI service, routing requests on your behalf without the calling application needing to know the details of what's behind it. The main problem it solves is quota management. Alibaba Cloud gives each registered AI model 10 million free tokens per account. Once that quota runs out, you would normally have to manually switch to a different account or model. This proxy automates that: it holds a pool of multiple API keys and multiple model IDs, and when one model's free quota is exhausted, it quietly switches to the next available model in the same account. If all models in an account are used up, it moves to the next account key. The application calling the proxy never needs to know any of this is happening. From the calling application's side, there is just one address and one API key. The proxy accepts requests in two common AI API formats: the Anthropic format and the OpenAI Chat Completions format. This means software already built to talk to either of those services can be pointed at this proxy with minimal changes. Real API keys are stored only on the machine running the proxy. The proxy never uploads keys anywhere, and the state file it uses to track cooldowns only stores a hashed version of each key, not the key itself. Setup involves copying a sample configuration file, filling in your real Alibaba Cloud API keys and model IDs, and running the server. A Docker option is also available for those who prefer container-based deployments. The README is written in Chinese.