← All explainers
Learn
Agents that stay accountable
An agent is a principal of its own with scopes and a budget. It acts only within the narrower of its own and its user’s permissions, destructive steps wait for a person, and every step is traced.
An identity, not a shared key
Each agent has its own principal, scopes, credentials and budget, and a registry of tools each declared with the permission it needs.
Never wider than its user
Under delegation an agent may do only what both it and the person running it may do, and never more than its own identity allows.
Limits that stop it
Allow-lists, spend caps, blast-radius limits, hard-stopped budgets and approval for destructive steps; a kill switch halts it within seconds.
A trace of every step
Every reasoning step and tool call is recorded with its decision, and the run can be replayed.