Gateway overview
Public-beta Gateway API — unified chat completions, models, and providers with fail-closed auth and dated pricing.
The Ethen Gateway is the developer control plane for model requests. Lifecycle for this release: public beta (not GA). Claims below match certification evidence and the portfolio registry.
How it works
- Authenticate with an Ethen Gateway API key (
Authorization: Bearer <key>). - Resolve the model against the Gateway catalog — or, for
ethen/*smart
aliases, resolve the alias to a concrete provider/model via the smart router (Smart routing).
- Enforce project scope, rate/concurrency/budget limits when configured.
- Call an upstream adapter (OpenAI, Anthropic, Vercel AI Gateway, …) using server credentials or caller-supplied keys where allowed.
- Return an OpenAI-compatible completion shape with an
ethenmetadata namespace. - Record usage against dated price records only — never OCR-only or expired prices.
Endpoints
| Endpoint | Method | Auth | Purpose |
|---|---|---|---|
/api/gateway/v1/chat/completions | POST | API key required | Chat completion (stream and non-stream) |
/api/gateway/v1/models | GET | Public catalog | List catalog models |
/api/gateway/v1/providers | GET | Public catalog | List provider metadata (no secrets) |
Authentication
Authorization: Bearer <ethen_gateway_api_key>Unauthenticated chat completion requests must fail closed (401/403 JSON). See Authentication and Gateway quickstart.
Provider readiness
Gateway launch paths with current certification receipts (as of M9 evidence):
- OpenAI
- Anthropic
- Vercel AI Gateway (upstream adapter)
Not Gateway launch-ready in this release documentation:
- Gemini (no Gateway certification receipt)
- Exa (Research support; not a Gateway launch provider)
- Ollama (local read-only certification only)
Details: Certification status and known limits.
Pricing honesty
Estimated cost uses only versioned official_provider price records with retrieval and expiry dates. Stale or OCR-only rows fail closed and must not be billed.