Capabilities
The AI gateway that shows its work.
One endpoint routes every request by cost, quality, latency and policy. Every call comes back with the receipt.
Decision traceresolved
- us-east / a100$0.42
- eu-west / h100$0.51
- us-west / l40s940 ms
- ap-south / a100quarantined
- route reason
- quality floor
- policy
- v14 eu-only
- cache
- miss
- overhead
- 3.1 ms
1endpoint
6signals scored per route
100%calls return a receipt
0silent fallbacks
What it does
Everything in one place.
Data plane
- OpenAI compatibleDrop in the base URL.
- Tool callingParallel calls, order kept.
- Structured outputJSON schema passthrough.
- MultimodalText, image, voice, video.
Routing
- Provider routingOne model, many endpoints.
- Latency awareMeasured, not advertised.
- Hedged requestsRace a stalled endpoint.
- CascadeEscalate only when needed.
- Semantic cacheStop paying twice.
- Region affinityStay close, stay compliant.
Evidence
- Routing receiptsWhy this route won.
- OTel exportInto your existing stack.
- Replayable decisionsPolicies are versioned.
- Cost reconciliationDown to the request.
Governance
- Bring your own keysYour provider contracts.
- Residency rulesEnforced before dispatch.
- Zero retention pathsEndpoints that keep nothing.
- Orgs and workspacesScoped keys and audit logs.
Why it holds up
Auto routing you can check.
Auditable
Candidates, constraints and scores come back with the response.
Trust layerCost controlled
Cheapest route that still clears the quality floor you set.
PricingQuality enforced
Providers are scored on measured behaviour, then quarantined or promoted.
Provider standardGoverned
Keys, budgets, residency and audit logs on one API.
Controls
Roadmap
In the architecture, not yet live.
Open to design partners. Not part of what you get on signup today.
- Quality floor SLAs
- Outcome based routing
- Learned router
- Batch and deadline market
- Capacity futures
- Verified compute