Working together in chat
The Chat tab is where most work happens: attachments, mentions, slash commands, live narration, and inline requests.
Open any employee and you land in Chat. It works like messaging a colleague who happens to be very fast and never sleeps.
Sending work
- Enter sends, Shift Enter adds a new line.
- Attach files with the plus button or just paste. Screenshots, PDFs, spreadsheets, docs, images, audio, video, and code files all work.
- Type @ to bring one of your other employees into the task, for a handoff or a second pair of hands.
- Type / to insert a connected app, an installed skill, or a surface like the browser or computer, telling the employee exactly what to use.
While they work
Employees narrate as they go: short milestone lines between the actual steps, like a colleague talking you through their screen. Tool activity shows as compact steps you can expand. If a run is going somewhere you don't like, the send button becomes a stop button while they're working.
Above the composer sits the autonomous status bar. It shows whether the employee is Working, Waiting for work, or Paused, and lets you turn Autonomous mode on or off without leaving the chat.
When they need you
Employees don't ask permission to act, so a card only appears when they genuinely can't get something themselves. Instead of failing quietly, they ask, inline in the chat:
- A reply box, when they're missing information only you have.
- A choice, when there are a few reasonable directions and they want you to pick.
- Connect an app, when the task needs a tool you haven't connected. The card has the connect button right there.
While a card is waiting, that employee goes quiet: no nudges, no status texts, no asking twice. It keeps working on anything it can finish without you, and once you answer it picks the blocked work back up and mentions what it got done meanwhile in one line.
These same requests are collected on the Runs page under Needs action, with a badge in the sidebar, so nothing waits on you invisibly. Unanswered requests expire after about three days and the run winds down safely. Details in Requests and Needs attention.
Giving feedback
Under each reply you can copy the text, try again, or rate it with a thumbs up or down. Ratings train the employee. So do plain words: “too formal, keep it to three sentences” changes future work, not just this reply. The Context button opens a drawer showing what went into a reply: your prompt, attachments, its reasoning, memory it drew on, and web sources.
Work you didn't start
Scheduled runs, triggered runs, and autonomous work stream into the same chat, live, exactly like work you asked for. You can watch an employee do its 6am inbox pass with your coffee, or scroll back through it later.
Tip
Corrections are the highest value thing you can type. One clear correction quietly becomes memory and improves every future run.