/ api

OpenAI-shaped. ORVEX-powered.

Swap the base URL. Keep the code. Get a distributed runtime, streaming out of the box, priced by the token served.

/ lifecycle

A single request, timed.

Every millisecond a request spends inside the runtime. This is the real shape.

request lifecycle · orvex-1 · us-west-2693 ms total
TLS
8ms
Auth
3ms
Route
2ms
Queue
4ms
Warm
6ms
Infer · first token
38ms
Stream (2.1k tok)
620ms
Settle
12ms
/ endpoints

All the surface, in one table.

POST/v1/chat/completionsChat, streaming, tools.
POST/v1/embeddingsBatched, deduped, cached.
POST/v1/images/generationsRouted to image-capable workers.
POST/v1/visionMulti-modal input, streaming output.
GET/v1/modelsLive catalog with capabilities.
GET/v1/workersNetwork-wide worker status.
GET/v1/receipts/:idSigned receipt for any request.
GET/v1/usageCredits, tokens, requests by window.
/ example

Streaming, in seconds.

A single POST. The response starts before the model finishes.

POST https://api.orvex.dev/v1/chat/completions
Authorization: Bearer $ORVEX_API_KEY
Content-Type: application/json

{
  "model": "orvex-1",
  "stream": true,
  "messages": [
    { "role": "user", "content": "explain the runtime" }
  ]
}
orvex · runtime · boot
booting
/ performance

Live, network-wide.

req · avg
–
4,820
TTFT · p50
–
38ms
stream · tok/s
–
182
errors · 5xx / 1k
–
0
target < 1
/ errors

Every failure, named.

One consistent error envelope. Every code has a documented meaning and a documented recovery.

400bad_requestThe request shape or values are invalid.
401unauthenticatedMissing or invalid API key.
402insufficient_creditsThe workspace is out of credits.
404not_foundModel or resource unknown.
429rate_limitedToo many requests. Backoff and retry.
500runtime_errorUs. Automatic retry on the network.
503no_worker_availableNo worker matched. Try a different model.
{
  "error": {
    "code": 429,
    "name": "rate_limited",
    "message": "Concurrency ceiling reached.",
    "retry_after_ms": 240,
    "request_id": "req_9f2a...",
    "worker": null
  }
}
/ rate limits

Predictable ceilings.

Developer
10 req / s
20 concurrent
Team
500 req / s
200 concurrent
Enterprise
custom
custom · private lanes