API Reference

The ORVYX public API is OpenAI-compatible. Every customer endpoint lives under a single base URL and speaks the same contract your OpenAI client already understands.

Base URL

https://api.orvyx-ai.com/v1

Authentication

All customer endpoints require an API key sent as a bearer token:

Authorization: Bearer oryx_live…

GET /v1/models and GET /health are the exceptions,they require no key.

Endpoints

Method	Path	Auth	Description
GET	/v1/models	none	List available models
POST	/v1/chat/completions	chat scope	Chat, streaming and non-streaming
POST	/v1/embeddings	embeddings scope	Text embeddings
GET	/health	none	Liveness / readiness probe

Every authenticated call travels the same path, which is why one client shape covers the whole catalogue.

1. 1. Your clientBearer oryx_…
2. 1. Key and scopechat, embeddings
   2. Route to a modelThe id you asked for
3. 1. OpenAI-shaped responseSame envelope as OpenAI

Authentication-related exclusions

GET /v1/models and GET /health are intentionally unauthenticated so they remain reachable during an incident. They expose only the public model catalogue and a process liveness signal,never account data.

Common headers

Header	Direction	Notes
Authorization	request	Bearer oryx_…
Content-Type	request	application/json on all POSTs
x-request-id	request (optional)	Client correlation id; echoed as x-client-request-id
x-request-id	response	Server-minted request id (always present)
X-RateLimit-Limit	response	Budget for the current window
X-RateLimit-Remaining	response	Remaining budget
X-RateLimit-Reset	response	Window reset indicator

Rate-limit headers are attached to all responses, not just rejections, so you can track budget proactively.

Response conventions

• Responses are normalized to the OpenAI schema.
• id, created and model are issued or rewritten by ORVYX. model always echoes the model id you requested.
• All timestamps are Unix seconds (UTC).

Errors

Every error uses one envelope, defined in Errors:

{
  "error": {
    "message": "…",
    "type": "invalid_request_error",
    "code": "invalid_request",
    "param": null,
    "request_id": "req_…"
  }
}

Pagination

The public API has no paginated endpoints,callers request one completion or one embedding batch at a time. Account-level resources such as usage history are available in the dashboard, not through the API.

Versioning

The /v1 prefix is the API version. The OpenAI contract is stable, so breaking changes are not introduced inside a version. Additive fields may appear at any time; treat unknown fields as forward-compatible.

Reference pages

• Chat Completions
• Embeddings
• Models
• Errors
• Rate Limits