AI Gateway

Route each step to the right model, and keep regulated data on your own infrastructure.

Run sensitive work on your own models and send everything else to the strongest frontier model, all in one workflow, without regulated data ever reaching a cloud. One gateway registers your local runtime and GPU cluster alongside the providers you already trust (Claude, GPT, Gemini, Azure), then routes each step by tier, cost, and data classification.

5Model providers, local + Claude/GPT/Gemini/Azure
3Routing tiers — local / GPU / governed cloud
per-stepModel choice in every workflow & agent
403Sovereign data blocked from cloud
How it routes

One gateway sends the right model to every step.

Register everything once, and tier, cost, and data classification then decide where each step runs, whether that is local, on your GPU cluster, or on a governed cloud model. Hybrid Claude, GPT, Gemini, and Azure calling is available today rather than a roadmap item.

Any model, any step

Never locked to one vendor

Bring any provider you like. Five come built in: a locked local runtime plus OpenAI, Anthropic (Claude), Gemini, and Azure OpenAI. Tier aliases (local-standard, local-advanced, cloud-reasoning) let you swap the model behind a step in configuration, with no redeploy.

Per-step routing

Run each step on the model it needs

The designer's Inference node and every agent step pick their own tier. A summarise step can run on a local model while a reasoning step in the same flow calls a governed cloud model, so you pay frontier rates only for the steps that need them.

Sovereign by default

Sovereign data stays on local tiers

Routing obeys the sovereignty gate. Sovereign-classified data can only reach local tiers, and any attempt to send it to the cloud returns 403 SOVEREIGNTY_BLOCK before it leaves your network. 'Auto' resolves to local first, so the safe path is the default.

Model-Ops

Add a model without a maintenance window

Catalog → VRAM fit-check → pull → activate, with multi-route failover on health cooldown so serving never drops mid-run. (Route changes are an Enterprise capability.)

Cost governance

See where every dollar goes

The gateway accounts for spend on every call, applies per-agent budgets, and attributes cost by route, so you can see exactly what each model and workflow costs. Admins can also enable per-user scoped-key budgets and rate limits.

Editions

Start routing on any edition

Model routing ships in every edition; Model-Ops route mutations are Enterprise.

See the gateway route on your own data.

Watch one workflow run a local model for sensitive steps and a governed cloud model for reasoning, with the sovereignty gate returning 403 the moment protected data heads for the cloud.

COMING SOONAANCER launches shortly.Register for prelaunch events & demos →