Control

Limits and safety

Agents are powerful, so you stay in control. Every agent acts with your access and never more, and every change waits for a person to approve it.

What an agent can do

  • It has the same permissions as the person who approved it, narrowed to what it asked for. It can never be given administrator access.
  • Permissions are checked on the server for every call.
  • Passwords and keys stay in your cloud account. Archon keeps a reference, and secrets are removed from what agents receive.
  • Changes go through preview and approval. An agent cannot approve its own change.

Your controls

On the Agents page you can, for each agent:

  • Limit it to some projects. Leave all unticked for every project you can reach.
  • Set a pace. Calls per minute and changes per day.
  • Pause it. It stops immediately. Pause all agents with one button.
  • See what it did. The latest 50 things it tried, including what was refused.
  • Disconnect it. Access ends at once. You can connect it again later.

If an agent keeps trying things it is not allowed to do (ten refusals in five minutes), Archon pauses it and tells you. Check what it did, then resume it.

The record

Archon keeps every change, who proposed it and who approved it. Agents are identified by the name of their connection, for example Claude Code, so you can tell your agents apart in the history. Records are kept when you disconnect an agent or a cloud account.

More detail

Permission names, the sign-in endpoints and token handling are in the reference.