A single endpoint. A network beneath it.
ORVEX is the layer between your app and the model. It picks the worker, streams the tokens, retries the failure, settles the payment and disappears.
What the runtime handles for you.
Everything below is on by default. Nothing is a plan tier. Nothing is a phone call.
Adaptive routing
Every call is scored against latency, price, load and locality. The best worker wins in under a millisecond.
Streaming natively
Token-level streams from the first byte. No polling. No queues you can see.
Model-agnostic
Open models, closed models, private fine-tunes. Same endpoint. Same billing.
Deterministic quotas
Credits, not surprises. Set a ceiling, get exact behavior at the ceiling.
Verifiable receipts
Every request produces a signed receipt: worker, region, tokens, cost. Auditable, ignorable.
Zero cold start
Warm pools per region keep TTFT under 50ms on production workloads.
See the runtime, always.
The dashboard is a live view of your traffic, not a weekly digest.