Get started
Govern one agent. Then decide.
The wrong first move is a platform decision. The right one is a single agent, in a sandbox, with one grant and one refusal you can point at — and the first two steps cost nothing.
- 1 · Read the vocabulary
- The specification is unbranded and about twenty minutes. If the entity model does not match how you think about your agents, nothing after this will fit either — and that is worth discovering before an integration.
- 2 · Open the demo
- A running deployment with real agents. Find a refusal, read what each gate did, and verify a pack from it in your own browser.
- 3 · Take a sandbox
- A private instance with synthetic data. Register your own agent, issue a grant against it, and watch it get refused for a reason you chose. Request one.
- 4 · Govern one real agent, in observe
- In your environment, with a grant that refuses nothing. A week of
observetells you what enforcement would have stopped, over your own traffic, before it stops anything. That report is the conversation to have internally — not a demo.
Most teams should not start with enforcement. Starting there means discovering your own edge cases as production incidents, which is how a control gets switched off and never switched back on.