AI Agent

Agent Tools

Pro+Agent Available on the Pro+Agent tier.

Tools are how the agent interacts with your workspace. When you ask “What pages mention Magistrate Thorne?” the agent doesn’t just guess — it calls search_pages, reads the results, and bases its response on what it finds. Tools make the agent useful rather than conversational.

How tool execution works

When the agent decides to use a tool, it assembles a JSON call with the tool name and parameters, sends it to the tool registry, and waits for the result. The result goes back into the agent’s context, and the agent uses it to formulate its response. This cycle can repeat multiple times in a single response — the agent might call search_pages, then call read_page on three of the results, and synthesize everything into a single reply.

The conversation panel shows tool activity as it happens: a badge with the tool name and “running…” appears below the thread, transitions to “done” when the call completes, and clears after two seconds.

Available workspace tools

The agent has access to these workspace tools by default:

Tool	What it does	Requires
`search_pages`	Full-text and semantic search across your workspace	PagesRead
`read_page`	Retrieve the full content and metadata of a page	PagesRead
`get_page_tree`	Get the full page hierarchy	PagesRead
`get_backlinks`	List pages that link to a given page	PagesRead
`get_outgoing_links`	List pages that a given page links to	PagesRead
`read_page_history`	Read event history for a page	PagesRead
`create_page`	Create a new page	PagesWrite
`update_page_content`	Update the content of an existing page	PagesWrite
`update_page_metadata`	Update page title, tags, or other metadata	PagesWrite
`move_page`	Move a page to a different location in the hierarchy	PagesOrganize
`rename_page`	Rename a page	PagesWrite
`delete_page`	Move a page to trash	PagesDelete

Tool result limits

Tool results are capped at 100 KB before being passed back to the agent. If a page is unusually large, the agent receives a partial result and will note that the content was truncated. The agent can still work with partial content — it will say so in its response.

Capability restrictions

Tools are gated by capabilities. The agent’s available capabilities are set in Settings → Agent → Capabilities. By default:

Read capabilities (PagesRead, SearchUse, TagsRead) are on
Write capabilities (PagesWrite, PagesDelete, PagesOrganize, TagsWrite, AttachmentsWrite) are off

When a write capability is off and the agent tries to use a write tool, it emits a permission request for you to approve or deny in the moment. You can also pre-grant capabilities in settings to skip the per-operation prompt.

Multi-step tool chains

The agent can chain multiple tool calls within a single session. If you ask it to “Read all three chapter pages and summarize them together,” it will call read_page three separate times and synthesize the results. Each call appears in the conversation panel in sequence.

The agent operates within a turn budget (20 turns by default). For most tasks, this is more than sufficient. If a session reaches the turn limit, the agent will report that it ran out of turns and provide whatever it completed.

Invalid and unavailable tools

If the agent hallucinates a tool name that doesn’t exist in the registry, the call returns a “tool not found” error. The agent receives this error as a tool result, acknowledges it, and continues without crashing. The conversation panel remains operational.