Frequently asked
Routing
Each provider has a cached profile covering quality on your task types, current latency and price per thousand tokens. Your rules set a quality floor and a budget; the router picks the cheapest profile that clears both. No extra model call is made to make that decision.
Median added latency is 31 ms, measured at the edge nearest your workspace. The scoring itself is sub-millisecond; the rest is network hops.
Yes. Pin a model on any call and routing steps aside, while failover, logging and spend tracking stay active.
Every decision is logged with the scores that produced it, so you can see exactly why a request went where it did and adjust the floor.
Reliability
Health checks run every second. A failing provider is pulled from the pool and the next-best model answers, typically within 40 ms, on an identical response schema.
Yes, with historical uptime per provider as well as for the routing layer itself. Incidents are posted before they are resolved, not after.
99.9% on Team and 99.99% on Scale, measured on the routing layer. Provider availability is reported separately so you can see which is which.
Cost
Never. You bring your own provider keys and pay them directly at your negotiated rate. We charge for routing only.
Set a cap per workspace, key or feature. As you approach it Zentic downgrades to the cheapest model that still clears your quality floor; at the cap it queues or returns a 429, whichever you configured.
One inbound call, however many providers it has to try. Failover retries are free and cache hits are not counted.
Security
Metadata lives in the region you choose. Prompt and completion bodies are dropped by default and only retained if you turn sampling on for evals.
No. Nothing that passes through Zentic is used to train any model, ours or a provider's, under any plan.
Both are available on request from security@zentic.io, along with completed SIG and CAIQ questionnaires.
SAML SSO and SCIM provisioning are included on the Scale plan.
Route your first request in about four minutes.
Free up to 50,000 requests a month. No card, no sales call, no migration.