Gateway API

The gateway is OpenAI-compatible. Point any OpenAI SDK at {ZEAGATE_PUBLIC_URL}/v1 with a zea- virtual key.

Method	Path	Auth	Purpose
POST	/v1/chat/completions	Bearer	Chat (streaming and non-streaming)
POST	/chat/completions	Bearer	Alias
POST	/v1/embeddings	Bearer	Embeddings (OpenAI / Azure only)
POST	/embeddings	Bearer	Alias
GET	/v1/models	Bearer	Model groups the key may use
GET	/health	None	DB connectivity
GET	/health/liveliness	None	Process up
GET	/health/readiness	None	DB + Redis

There are no other public HTTP routes. Cache invalidation is Redis pub/sub, not HTTP.

Request lifecycle (chat)

1. Auth — hash lookup; blocked/expired keys rejected
2. Model access — key allowlist
3. Budgets — every level must pass
4. Rate limits — RPM/TPM for key, team, application, customer
5. Pre-request guardrails
6. Routing — weighted pick, retries, fallbacks
7. Stream or return the provider response
8. Post-response guardrails
9. Async spend log

Error envelope

Gateway-generated errors:

{
  "error": {
    "message": "human-readable message",
    "type": "authentication_error",
    "code": "invalid_api_key"
  }
}

Upstream provider errors are passed through with the upstream status and body.

See Error codes, headers and SDK snippets.