AxonHub is a self-hosted AI gateway that lets an application switch model providers without code changes. Whatever SDK the client uses (OpenAI, Anthropic, Gemini, or another), AxonHub translates the request into the format the chosen provider expects, so you can call Claude through the OpenAI SDK or GPT through the Anthropic SDK by changing configuration only.
It targets four problems: provider lock-in, the work of integrating many provider APIs, missing request observability, and cost control. Requests are traced end to end with thread-level timelines, load balancing fails over to the healthiest channel in under 100 ms, role-based access control adds quotas and data isolation, and every request is costed by input, output, and cached tokens. An admin console covers channels, models, model pricing, traces, and request monitoring.
Supported API types are text generation (OpenAI, Anthropic, and Gemini formats), image generation, embeddings, and rerank, with realtime conversation planned. Providers and the API formats each accepts:
| Provider | Example models | Compatible APIs |
|---|---|---|
| OpenAI | GPT-4, GPT-4o, GPT-5 | OpenAI, Anthropic, Gemini, embeddings, image generation |
| Anthropic | Claude 3.5, Claude 3.0 | OpenAI, Anthropic, Gemini |
| Gemini | Gemini 2.5 | OpenAI, Anthropic, Gemini, image generation |
| DeepSeek | DeepSeek-V3.1 | OpenAI, Anthropic, Gemini |
| Zhipu AI | GLM-4.5 | OpenAI, Anthropic, Gemini |
| Moonshot | kimi-k2 | OpenAI, Anthropic, Gemini |
| ByteDance Doubao | doubao-1.6 | OpenAI, Anthropic, Gemini, image generation |
| Jina AI | Embeddings, rerankers | Jina embedding and rerank |
| OpenRouter, NanoGPT | Many models | OpenAI, Anthropic, Gemini, image generation |
| AWS Bedrock, Google Cloud (testing) | Claude | OpenAI, Anthropic, Gemini |
Deployment is handed to an agent: a companion agent skill walks a coding agent through setting up AxonHub, after which an initialization wizard creates the admin account and you add provider channels and client API keys.
Features
- Any SDK to any model: translates OpenAI, Anthropic, and Gemini SDK requests to any supported provider
- Request tracing: full request timelines with thread-level observability
- RBAC: fine-grained access control, usage quotas, and data isolation
- Load balancing: automatic failover in under 100 ms, routing to the healthiest channel
- Cost tracking: per-request cost breakdown across input, output, and cached tokens
- Multiple API types: text generation, image generation, embeddings, and rerank
- Admin console: dashboards for channels, models, pricing, traces, and request monitoring
- Agent-driven deployment: a companion skill guides a coding agent through deploying the gateway
