Skip to content

Security overview

Agent limits what workers can do with roles, policies, approvals, budgets and the kill switch, checks every outgoing call for secrets, and records everything in a tamper-evident log.

Controls on workersLink to Controls on workers

  • Roles and capabilities: a worker can use only the capabilities its role and settings allow. A denied capability always wins.
  • Risk tiers and policies: risky actions ask a person or are denied. See Risk tiers.
  • Approvals and the Always ask list. See Approvals.
  • Budgets: Agent reserves spend before each call. See Budgets and spend limits.
  • The kill switch stops everything at once. See Kill switch and pause.
  • Oversight watches for loops, spend spikes and unusual behaviour. See Oversight.

No setting weakens the hard gates: the kill switch, prohibited risk, the outgoing check and separation of duties.

The outgoing checkLink to The outgoing check

Agent checks every outgoing call. It blocks secret patterns, values from the work's environment and its own secrets. It holds for a person anything that looks like data sent out in a hidden way.

Outside contentLink to Outside content

Content from outside, such as a web page or an email, raises the risk of the action that follows it.

The recordLink to The record

The Audit Log records every change and can be verified. See Audit log.

LimitsLink to Limits

These controls lower risk. They do not make work risk-free. Check the evidence of important work.

Checked against the product on 2026-10-05.

Was this helpful?