An open-source control plane for AI API chaos

What it does Claude Code Hub is a self-hosted gateway that sits between your team and a mix of LLM providers—Claude, Codex, Gemini, or any OpenAI-compatible endpoint. It handles routing, circuit-breaker failover, usage metering, and rate limiting through a Hono proxy pipeline backed by PostgreSQL and Redis, with a Next.js 15 dashboard for live monitoring and cost attribution.

The interesting bit The README opens with six affiliate API-reseller promotions before the first feature list, which tells you exactly who this is built for: teams that need to hop between regional relay providers and official APIs without changing client code. It treats rate limits and price arbitrage as first-class problems.

Key highlights

• Weighted, priority-based load balancing with circuit breakers and up to three automatic retries across provider groups
• Granular rate limiting (RPM, spend caps, concurrent sessions) enforced via Redis Lua scripts with fail-open degradation
• Auto-generated OpenAPI 3.1 docs and Swagger/Scalar UI from 39 internal Server Actions endpoints
• Five-minute session context cache that records full decision chains for audit and cost attribution
• Strict OpenAI-compatible /v1/chat/completions routing without cross-format translation

Caveats

• Documentation is heavily interleaved with affiliate promotions for external API resellers
• Architecture details and advanced configuration are truncated in the provided README
• UI and primary documentation are Chinese-first, which may add friction for English-only operators

Verdict Grab it if you manage multiple provider keys and need a single internal endpoint with spend tracking and automatic failover. Pass if you already have a cloud-native gateway or just want a thin reverse proxy.