Chat

Ask in plain language and the agent works through your tools, skills, projects and connected apps, showing every step it takes along the way.

The agent beside every page

Open the agent next to any page and it already knows the file, table or app you are looking at.

Work on a document, a table or an app with the agent beside it, without copying text into a chat or explaining where anything lives.

The agent writes and runs its own code in a safe sandbox, so totals and data transforms are computed rather than estimated.

One script can read your files and tables, call your connected tools and save what it produced, so a multi-step job runs in a single go.

Every action the agent takes shows in the reply, with what it read and what it sent, folded into one line once it finishes.

Check what the agent read and sent before you trust an answer, which matters once it works in your team's email, files and tables.

Typing a slash lists your skills, each with its description and the argument it takes, and runs the one you pick.

An @mention points the agent at a file or a project to work on, and a # loads a whole project into the conversation.

Fast, Auto or Slow sits in the composer and sticks to that chat, so a quick question never waits on the slowest model.

Download one conversation as markdown or JSON, copy the whole transcript to the clipboard, or take every chat you own as one zip.

The sidebar and the Chats page put chats waiting on your answer first, then the ones the agent is working on, then the rest.

Start a chat that stays out of your history and the agent's memory, then deletes itself 30 days after the last message.

A long chat opens on its most recent messages, keeps your question pinned while the reply grows, and summarizes older history.