Trust mode
Trust mode sets which actions the agent takes on its own and which require your approval. Change it below the message field or in Settings.
Read-only
The agent reads and analyzes data. It asks before every command or edit.
Auto
The agent runs local commands and edits on its own. It still asks before actions in connected services and other risky actions.
Important actions stay under your control
Money-related and irreversible actions need approval in every mode.
When your organization picks the mode
In a work organization the administrator sets an autonomy ceiling — the freest mode members are allowed to use.
- Anything stricter than the ceiling is yours to choose: you can always hold the agent back further.
- Anything above the ceiling is unavailable: the option is labelled Set by administrator and cannot be selected.
- If your saved mode ends up above a new ceiling, it drops to the highest allowed level and you see the notice Adjusted by organization policy. Your choice is not lost: once the administrator raises the ceiling, your mode comes back.
An administrator can also forbid changing and irreversible actions in specific services. That ban outranks trust mode: even in auto the agent will not perform the forbidden action and says so in chat.
Frequently asked questions
How is read-only different from auto?
In read-only the agent reads and analyzes data and asks before every command or edit. In auto it runs local commands and edits on its own, while actions in connected services and other risky steps still wait for your approval.
Why is auto greyed out and labelled "Set by administrator"?
Your organization has set an autonomy ceiling. You can always pick a stricter mode than the ceiling, never a freer one; if your saved mode was above it, it drops to the highest allowed level. Only an organization administrator can raise the ceiling.