Faster AI adoption with control over agent actions

14 days, counted from the day your tenant is set up. It is your own tenant, not a shared sandbox, and the trial is free.

It sits inline: every tool call passes through the gateway and gets an allow or deny decision before it runs. A new tenant starts with a single Allow all rule, so in practice you begin observe-first: agents keep working, every call is recorded in the Agent Trail, and you tighten the rules once you have seen what your agents actually reach. There is no mirror-only mode where calls bypass the gateway.

For each call it records structured metadata: the timestamp, the user and agent instance, the action, the resource and tool, the policy decision, and the request context. It does not store the prompts or responses of the agent's conversation with its model. Tool responses pass through unchanged unless you switch on a Data Protection filter, and a redacted call records only which rule matched and how many values it replaced, never the values themselves. Audit events are kept for three months, and webhooks can forward them to your own systems.