Agent Permissions

Pro+Agent Available on the Pro+Agent tier.

The agent can read your workspace, search your pages, and — with your approval — create, edit, and delete content. Permissions are capability-based: you control which categories of operations the agent can perform, and the agent cannot act outside those boundaries.

How permissions work

The agent declares the capability it needs before executing any write operation. If that capability hasn’t been pre-granted in settings, the agent emits a permission request and pauses until you respond.

This means you’re always in control of what changes in your workspace. The agent asks; you decide.

Permission request dialog

When the agent needs a capability, a permission dialog appears inline in the conversation thread. It shows:

• The tool name the agent is trying to use (in monospace, e.g., create_page)
• The capability required (e.g., pages_write)
• An explanation from the agent describing why it needs this capability

Three buttons are available:

Option	Effect
Allow Once	Grants permission for this specific tool call only. The next request for the same capability will prompt again.
Always Allow	Grants permission for the remainder of the current session. The agent proceeds without interrupting you for the same capability again.
Deny	Blocks the tool call. The agent receives the denial and responds accordingly — typically explaining that it couldn’t complete the requested action.

Buttons are disabled while the response is in-flight to prevent duplicate submissions.

Permission banner

If the conversation panel is closed when a permission request arrives, a banner appears at the bottom of the screen. The banner shows the required capability and two controls:

• View — opens the conversation panel where the full dialog is shown
• X (dismiss) — hides the banner without denying the request

Dismissing the banner is not the same as denying. The permission request remains pending in the conversation thread. Opening the panel later reveals the dialog, still awaiting your response.

Pre-configured capabilities

You can grant capabilities permanently from Settings → Agent → Capabilities rather than responding to per-operation requests. Default capability settings:

Capability	Default	What it allows
PagesRead	On	Read page content and metadata
SearchUse	On	Search your workspace
TagsRead	On	Read tag information
PagesWrite	Off	Create and edit pages
PagesDelete	Off	Move pages to trash
PagesOrganize	Off	Move pages in the hierarchy
TagsWrite	Off	Create and edit tags
AttachmentsWrite	Off	Upload attachments

Turning a capability on in settings means the agent can use it without asking. Turning it off means the agent will always ask — or won’t ask and will report it can’t do something if you’ve denied that capability’s runtime requests.

Capability isolation

Different capabilities are independent. Granting pages_write does not grant tags_write. If you allow a page creation and then ask the agent to tag the page, it will issue a separate permission request for tags_write.

Caution

Pre-granting write capabilities means the agent can modify your workspace without asking each time. This is convenient for heavy usage, but means mistakes (accidental deletions, unexpected edits) require manual recovery rather than a simple “Deny.” Start with per-operation approval and move to pre-grants as you build trust with specific workflows.

See Also

• Agent Conversation — How permission dialogs appear in the thread
• Agent Tools — Which capabilities each tool requires
• Getting Started with Agent — Start the agent
• Configure LLM Providers — Set up your API key