You set what the agents may do, and you can check any step they took.
A person sets what each agent may do with each tool. Governance rules block or pause named tool calls and scan every incoming message for prompt injection. Every step and tool call is logged, and the log can be sent to your own monitoring tools.Last checked 27 October 2026
| Control | What it does |
|---|---|
| Approvals | Anything a customer sees, money spent or records changed waits for a person, by default. Several people can be approvers, and the first decision wins. |
| Rules | Block or pause any tool call, whatever the model decides. Messages are scanned for prompt injection before the agent reads them. |
| Permissions | A person sets what each agent may do with each tool. |
| Roles | Viewer, operator, builder, admin, owner and billing admin, each with a fixed set of rights. Only an owner can grant the owner role. |
| Run history | Every step and tool call is recorded and can be sent to your own HTTPS endpoint, OpenTelemetry, Datadog, Better Stack, Splunk, Grafana Loki, Axiom, New Relic, Sumo Logic or object storage. |
| Browser | An agent that uses a website is locked to the sites you allow, and anything that buys, pays, signs in or sends waits for a person. |
| Spend | Each agent team has its own pool of actions, and every reply shows what it cost. |
Compliance and data handling
Ask us for the current report and our data handling summary. The list of subprocessors and the privacy notice are public.
Common questions
Can I control what each agent can do?
Yes. Approvals are built in, governance rules can block or pause any tool call whatever the model decides, and a person sets what each agent may do with each tool.
Is every step recorded?
Yes. Every step is logged and can be sent to your own monitoring tools.
Where can I find compliance details?
Ask us for the current report and data handling summary.