The agent

Open the agent

Select Agent in the sidebar, or press Shift+A on any workspace page. Press Shift+A again to close it. The agent keeps its conversation while you move between pages.

Choose where it sits in the … menu: Docked on the right, Floating, bottom right, or Floating, bottom left. You can also drag its title bar to one of these places. Docked, the app moves aside and the agent works next to it. On a phone it opens full screen. A floating agent can be minimized to a bar in its corner.

Conversations you start here are in your Personal inbox. Only you can read them. A new conversation lists what waits for you and your recent conversations; the clock button lists all of them. Open in Inbox in the … menu moves the conversation to the full page. When you leave the Inbox with one of your own conversations open, it continues in the agent, so you can keep working on it anywhere in the app.

Ask about what you are looking at

On an asset, work order, or space, the composer shows that object beside the attach button. The agent knows you mean it. Remove it with × if you mean something else. A new conversation also suggests questions about it.

Attach files with the paper clip, or paste or drop them into the composer, such as a screenshot or a photo.

What the agent can do

The agent works with your permissions. It can:

• search the workspace, and list assets and work orders by type, space, status, people, and dates, for example "work orders changed in the last 4 weeks";
• read assets, work orders, spaces, parts, companies, contacts, records, and files, including spreadsheets, Word, PowerPoint, CSV, and text;
• create and change assets, spaces, work orders, and records, such as a note, photos, time, or costs on an asset;
• write a work order's document: add, change, or remove steps with text, tasks, inspections, choices, and tables, or insert a template;
• add parts to the parts catalog with their manufacturer, suppliers, and the asset types they fit, and create companies;
• set up asset types and properties, and save a list as a view;
• invite contacts and members;
• make one change to up to 50 assets or work orders at once, for example set a status or add a tag;
• move things to Trash and restore them.

When the agent has no way to do what you asked, it says so and where in the app you can do it. It does not save it as something else. When it does only part of a request, its reply says which part it left out and why.

In a work order's document, the agent leaves images and other blocks as they are. If someone is editing the document at the same time, both edits are kept.

You can name things the way you know them: a work order number such as MAINT-12, an identifier or serial number, a name, or a link.

The agent shows each step as it works, such as "Searching" while it runs and "Searched" when it is done. Select a step to see what it sent and what came back. When the agent is done, its read steps fold into one line above the reply; the changes it made stay in view.

Changes ask first

By default, the agent shows each change in a box, with the fields before and after, and waits:

• Approve makes the changes.
• Reject discards them. The agent reads your decision and continues.
• Approve and allow later changes in this conversation, in the menu beside Approve, approves and lets the agent change data without asking for the rest of the conversation.

Once you decide, the buttons give way to your decision. A change the agent makes without asking shows in the same box.

Writing a message while the agent waits rejects the changes and tells the agent what to do instead. Turn off Ask before changes in the … menu to allow changes in a conversation.

Moving something to Trash, restoring it, and inviting people always ask. So does a conversation with someone from outside the workspace.

Changes the agent makes show in activity and records as Agent on behalf of your name, linked to the conversation.

Questions, stopping, and taking over

• When the agent asks a question, answer with one of its buttons, in the composer, or by email.
• While the agent works, a new message adds an instruction to the current work.
• Stop takes the place of Send while the agent works and ends the current work at once.
• Take over, in a shared conversation's toolbar on the Inbox page, stops the agent so you answer yourself. Return to agent hands it back.

Closing the page does not stop the agent. The sidebar shows when it works for you and how many conversations wait for your answer or approval.

Limits

A conversation starts at most 10 agent runs an hour, and one run makes at most 25 steps. A workspace has 1,000 agent steps a day. A run ends after seven days at the latest.

Images and PDFs reach the agent when the configured model supports them. See Limits and file types.

Without an AI provider

If no AI model is configured, the agent replies that it cannot answer yet and that your message is saved. Owners and admins choose models in Workspace settings, AI models.

Who can

Action	Who
Use the agent	Members
Approve the agent's changes	The member the agent works for
Answer, stop, take over, or return	Members assigned to the inbox
Set whether the agent asks before changes	Owners and admins
Choose the workspace's AI models	Owners and admins