The right panel is a chat with Seben’s AI agent, which can read and edit the open document for you. Try things like:
- “Summarize this PDF”
- “Fix typos throughout”
- “Redact all emails”
- “Fill this form”
As the agent works, it shows each tool step it takes, so you can see exactly what it is doing to your document as it happens.
When text alone does not answer a question, the agent can look at a page as an image: a chart, a table’s layout, a stamp or handwriting. It looks at no more than four pages per answer, and the images are not kept with the chat.
On a scanned page, the agent reads the text OCR recognized and is told that it came from a scan.
Ask it to “prepare this for release” and it starts a release run and gives you a link to review it. The agent cannot approve a release; you review and approve it yourself.
It can also start the other playbooks: put documents into a binder, compare two drafts, or prepare contracts for signature. A binder or comparison is created when the run finishes; contract runs wait for you to review and approve them.
Ask mode and Auto mode
The Ask and Auto switch at the top of the editor decides which changes wait for you.
- Ask is the default. The agent waits for your approval before deleting pages, redacting, reordering, merging, splitting or digitally signing. It shows an approval card with Approve and Reject, and nothing in that request is applied until you approve it. Other changes, such as fixing text or filling a form, are applied as the agent works.
- Auto applies those changes too, without asking. While it is on, the panel says “Working autonomously”.
In either mode, the agent asks before every call to a connected app, and every change it makes is a new version you can undo. File → Version history… lists them all.
Choosing a model
Use the menu at the top of the agent panel to choose which AI model the agent uses.