BenedictKing/ccx
Ccx
CCX is a self-hosted AI API gateway that puts Claude, OpenAI, and Gemini behind a single endpoint and credential. It translates between the providers' protocols, so clients can mix upstream services through whichever API format they already use. It accepts Claude Messages calls, OpenAI-style chat and Codex requests, and Gemini content-generation calls, and it also handles embeddings, model listing, and image generation and editing.
Requests are scheduled across configured upstream channels by priority, with promotion windows that decide when a preferred channel gets traffic again. Continuous health checks skip failing channels and restore them once they recover. Routing also filters channels by each model's real context window and output size, and multi-turn conversations stay on the channel that handled the earlier turns. Each channel can rotate through several API keys and can have its own proxy, custom headers, model allowlists, model name mappings, and route prefix.
A built-in web console manages channels and keys, reorders priorities by drag and drop, and shows logs and live per-channel request volume, success rate, and latency. It also shows model benchmark rankings and pricing and context-window data refreshed from a public community-maintained registry. The proxy is guarded by an access key, with an optional separate admin key, and the gateway can bind to localhost only or listen on all interfaces.
The Go backend embeds the compiled frontend, so one process and one port serve both the console and the proxy. It runs as a downloaded binary, in Docker or Docker Compose, from a source build, or as a desktop app on Windows, macOS, and Linux. Behavior is configured through environment variables, and a health endpoint reports whether the gateway is running.
