The universal translator for AI SDKs

What it does

AxonHub is a self-hosted Go gateway that sits between your application and a dozen LLM providers. You keep your existing SDK code—OpenAI, Anthropic, Gemini, whatever—and point it at AxonHub instead of the vendor’s endpoint. The gateway translates requests, routes them to the configured provider, and handles failover, load balancing, and cost tracking.

The interesting bit

The bidirectional SDK compatibility is the real trick. Most gateways standardize on the OpenAI API format and force you to adapt; AxonHub claims to let you use any supported SDK to call any supported model. Whether the translation is equally complete across all SDK-to-model combinations is unclear from the README, but the ambition is notable.

Key highlights

• Supports 10+ providers including OpenAI, Anthropic, DeepSeek, Gemini, Moonshot, ByteDance Doubao, and OpenRouter
• Auto failover in <100ms with health-check-based routing
• Per-request cost breakdown tracking input, output, and cache tokens
• Built-in RBAC, usage quotas, and request tracing with thread-aware timelines
• Self-contained binary with SQLite default; no external dependencies required

Caveats

• Realtime/live conversation API is marked as TODO
• AWS Bedrock and Google Cloud integrations are still in testing
• The “<100ms failover” claim lacks methodology or benchmark details in the README

Verdict

Worth evaluating if you’re running multiple AI services and tired of maintaining separate provider integrations. Less compelling if you’re already committed to one ecosystem and don’t need the abstraction overhead.