wyverncw/freeproxy

FreeProxy (called OpenCode Proxy in the README) is a small server you host yourself that acts as a bridge between AI chat apps like JanitorAI and the OpenCode AI service. JanitorAI is a roleplay chat app that lets users bring their own AI backend, and this proxy makes it possible to use OpenCode's free AI models with that app without paying for a commercial API tier. The proxy works by accepting requests in the standard OpenAI-compatible format, the same format used by ChatGPT's API. When JanitorAI or another client sends a chat message to the proxy, the proxy forwards it to OpenCode and returns the response. This means any chat app that supports the OpenAI API format can use this proxy to access the free OpenCode models. One practical feature is multi-key failover. You can provide up to three OpenCode API keys, and if one hits a rate limit, the proxy automatically switches to the next one. Responses stream in real time using a technique called Server-Sent Events, so text appears progressively rather than arriving all at once after a delay. The proxy includes several security measures by default. It limits each IP address to 100 requests per minute, validates the size and structure of incoming requests, and sets standard security headers on all responses. It stores no logs of any kind, so neither request content nor user metadata is retained anywhere. Deployment is designed around Vercel, a hosting platform where you can run small servers for free. You clone the repo, install dependencies, and run a single deploy command. Then you add your OpenCode API keys as environment variables in the Vercel dashboard and point JanitorAI's settings at your proxy URL. Four free AI models are supported: mimo-v2.5-free, deepseek-v4-flash-free, nemotron-3-ultra-free, and north-mini-code-free. The project is MIT-licensed.