One port, five AI APIs, a web UI, and a prayer

What it does

CCX sits between your application and five different AI provider APIs—Claude Messages, OpenAI Chat Completions, OpenAI Images, Codex Responses, and Gemini—translating and routing requests through one local port. It bundles a web admin interface, channel orchestration, failover logic, and per-channel key rotation into a single Go binary with embedded frontend assets.

The interesting bit

The “channel orchestration” concept: you configure multiple upstream providers with priorities, health checks, and promotion windows, then drag them around a visual dashboard to control traffic flow. When one provider fails, the circuit recovers automatically. It’s the kind of operational tooling most API proxies leave as a YAML exercise.

Key highlights

• Single-port deployment serves both the proxy endpoints (/v1/chat/completions, /v1/messages, etc.) and the web UI (/)
• Dual-key auth separates proxy access from admin access
• Per-channel controls: API key rotation, custom headers, model allowlists, route prefixes, and proxy settings
• Session tracking for Codex Responses multi-turn workflows
• Desktop app available via Microsoft Store and Homebrew Cask

Caveats

• Docker image is hosted on a personal Alibaba Cloud registry (crpi-...aliyuncs.com), which may raise reliability or trust questions
• Windows networking quirks explicitly documented: localhost may not resolve from WSL/Docker, requiring raw IPv4
• Go 1.25+ and Bun required for source builds—bleeding-edge toolchain

Verdict

Worth a look if you’re running multiple AI provider keys in production and tired of hand-rolling failover logic. Overkill for personal API key management; the visual dashboard is the selling point, not the proxying itself.