Zero tokens while it waits
It waits with no model running. In Claude Code and Codex it checks in about once a day with one short turn.
Autopilot
You step away.
Your agent doesn’t.
Turn on autopilot, and a trusted teammate's work request goes straight to your open agent session. It does the task, then waits for the next one. On Team and up.
How it works
Turn it on, leave your session open, come back to finished work.
Pick who can give it work. Everyone in your org by default, including people who join later, or only the people you choose. Keep it to one repository or all of them. Anyone else still waits for your yes.
Who can give your agent work →
Requests start right away. A trusted teammate's request is accepted for you, and your open session picks it up in the same chat. One task at a time, then it waits again with no model running.
How it works →
When you're back, it tells you. Your next prompt turns autopilot off, and the agent shows what it did first. The same recap goes to the dashboard feed and to your Slack channel.
Turning it off →
The details
What it does while you're away, and how you stay in charge.
Your session, not a bot
No extra program, no background agent. The same chat, repo, setup and permissions you already use.
Coming back turns it off
Press Esc and type, run /autopilot off, or flip the switch in the dashboard. The recap comes first.
Summary when you're back
A separate switch on every plan: after 8 hours away, your next session opens with what happened in your repos.
Claude Code, Codex and omp
Autopilot runs in all three. In other agents, requests to you stay requests and wait for your yes.
On the record
Turning it on or off, and every summary, goes to the audit log and to your webhooks.