Daily use
Models and permission modes
Choose the model and the permission level per session: plan, read-only, workspace write, and full access.
Choosing a model
Each session can pick its own model. Models differ in reasoning depth, speed and tool-calling ability — switch per task rather than setting one globally.
Permission modes
The permission mode decides what the agent can do, not merely whether it asks:
| Mode | Allowed | Good for |
|---|---|---|
| Plan | read-only exploration, produce a plan | look before you leap |
| Read-only | read files, search, browse | research and questions |
| Workspace write | edit files and run commands inside the workspace | everyday changes |
| Full access | step outside the workspace, broader system actions | when you know exactly what you want |
Grant, then take back
Permissions are granted per session. Do not enable writes for read-only research, and lower the level when the task is done — that is far safer than leaving full access on.
Approval points
Even in a high-permission mode, irreversible actions — sending, deleting, buying, publishing — are explained before they run.
Next
- Permissions — the default-deny policy for phones and plugins.
- Auditability — how to review what it did.