Multi-tenant AI gateway
AI access, simplified.
One API. Every model. No complexity — route, meter, and govern every provider your team uses from a single control plane.
Overview
| Model | Provider | Requests | Cost / 1M |
|---|---|---|---|
| gpt-4o | openai | 412k | $5.00 |
| claude-sonnet-4 | anthropic | 308k | $3.00 |
| gemini-2.0-flash | 255k | $0.15 | |
| llama-3.1-70b | meta | 189k | $0.60 |
Every major model, one endpoint
Built for platform teams
One gateway between your product and every provider.
Unified API surface
Point your SDK at one base URL. The gateway translates requests and streamed responses across providers, so switching models is a string change — not a rewrite.
Multi-tenant by design
Issue per-workspace keys, set budgets and rate limits per tenant, and isolate usage so one customer's spike never lands on another's bill or quota.
Metering and cost control
Every request is logged with tokens, latency, and price. Attribute spend to a workspace, key, or end user, and fail closed when a budget is exhausted.
How it works
Live in three steps.
No new infrastructure to run. The gateway is hosted — you bring provider keys, we handle routing, metering, and failover.
Connect your providers
Add provider keys once. They are encrypted at rest and never exposed to tenant workspaces or client code.
Issue keys per tenant
Create a workspace for each customer, set its budget and rate limit, and hand out a scoped key. Permissions and quotas are enforced at the edge.
Route and observe
Call any model through the unified API. Watch usage, spend, and errors per tenant in the dashboard, with retries and fallback models when a provider degrades.
The control plane
Govern access without slowing developers down.
Central policy, per-tenant enforcement. The gateway is the one place a key is checked, a budget is counted, and a request is logged.
- Per-workspace budgets — hard and soft caps, reset monthly or rolling.
- Rate limits and retries — token- and request-based limits with automatic fallback models.
- Full request logs — model, tokens, latency, cost, and status for every call.
- Bring your own keys — encrypted at rest, scoped per provider, auditable on rotation.
Pricing
Pay for requests, not seats.
Every plan includes the full gateway. Plans differ by workspaces, log retention, and support.
| Included | Free | Pro | Scale |
|---|---|---|---|
| Workspaces (tenants) | 1 | 10 | Unlimited |
| Provider integrations | All | All | All |
| Request log retention | 7 days | 90 days | Custom |
| Per-workspace budgets & limits | — | ✓ | ✓ |
| SSO & audit export | — | — | ✓ |
| Monthly | $0 | $49 | Talk to us |
We stopped rebuilding our provider layer every time a model shipped. Teams pick a model in the dashboard and it just routes.
FAQ
Questions, answered.
Which providers are supported?
OpenAI, Anthropic, Google, Meta, Mistral, and Cohere at launch, with an OpenAI-compatible API for anything else. New providers are added without changes to your integration.
Where do my provider keys live?
Keys are encrypted at rest and used only server-side by the gateway. They are never returned to tenant workspaces or exposed to client applications, and every use is logged.
How does multi-tenancy work?
Each tenant is a workspace with its own keys, budgets, rate limits, and usage history. You can cap spend per workspace and attribute every request to a workspace, key, or end user.
What happens when a provider goes down?
Requests can be configured with a fallback model or provider. If the primary fails, the gateway retries against the fallback and records the switch in your logs.
Is there a markup on provider usage?
No. You pay the provider's published rate, passed through at cost. The plan price covers the gateway, dashboard, and support.
Ship with one API, not ten.
Point your SDK at the gateway and route to any model. Free to start — no credit card.
Get Started