# Matimo Gateway

Gateway is how you govern agents you built outside Workbench. Point an
OpenAI-compatible or Anthropic-compatible client at Gateway instead of the
provider by changing its base URL. Every model call is then identified, checked,
routed and recorded.

## What it does

- **Identity.** Each registered agent has its own cryptographic identity, so
  calls are attributable.
- **Checks on every call.** Your Guardrails and policies run before the request
  reaches the model.
- **Your keys stay put.** Gateway uses your own provider credentials on the
  platform side. They never sit in the agent's process.
- **Spend control.** Per-agent spend and token caps are reserved before a call
  is sent. Virtual keys carry budgets, rate limits, allowed models and expiry.
- **Tool checks.** A tool call can be allowed, denied, or held for human
  approval.
- **Observability.** A filterable call log and a run timeline show what each
  agent did and what it cost.
- **Telemetry.** Agents report activity, and an agent that goes quiet is flagged
  as absent, not just unobserved.
- **One record.** An agent that registers once keeps the same identity, policy
  history and audit trail whether it uses Gateway directly or through
  [AGDK](/docs/agdk).

## What it does not do

- It does not publish a latency figure.
- It does not enforce tokens-per-minute limits.
- There are no ready-made plugins for individual agent frameworks. Integration
  is by base URL and the Gateway API, and your engineers do the wiring.

## Availability

Available with Matimo Enterprise.

## Who it is for

Teams already running agents on LangChain, CrewAI, Google ADK, AutoGen or custom
code who need identity, policy and an audit trail without rebuilding them.

## Next step

[Contact us](/contact) or write to info@roaiq.com to scope an integration.