Overview
Overview
ATP is a unified API that gives you access to many AI models through a single endpoint, while handling provider fallbacks and billing in one place. Point any OpenAI-, Anthropic-, or Gemini-style client at the Gateway and pay for usage in credits.
What you get
- One endpoint, many models. Discover them with GET /v1/models and call any your project allows.
- Three wire formats. Use the OpenAI, Anthropic, or Google GenAI shape unchanged — only the base URL changes.
- Automatic fallback. Each model is served by a provider pool, so a single model id keeps working when one provider is degraded. See Provider routing.
- One bill. Usage across every model and provider is metered in credits. See How credits work.
Three ways to integrate
| Approach | Best for |
|---|---|
| API | Full control, any language, no dependencies |
| SDKs | Type-safe calls with your existing OpenAI / Anthropic / Google SDK |
| Coding agents | Claude Code, Codex, and other agents that speak the wire format |
Start here
New to ATP? Run through the Quickstart, then read How it works to understand the request lifecycle. When you're ready to organize keys and budgets, see Set up your organization.