Agents that start themselves — on a schedule, or an event.
An automation wakes an agent under permissions you set: a push, an alert, a payment, an email, a chat message, or plain cron. A guard command you write decides whether each wake runs, and every run is a fresh session you can watch.
Its own transcript, its own worktree, the same review before anything lands.
- 6 eventspush, alert, payment, email, chat or cron
- 1 guarda command you write vets every wake
- Fresh sessioneach run its own transcript and worktree
A persona with permissions, not a free-for-all
Each automation runs as an agent persona you've configured — which model, which tools, how much it may do unattended. The guard command runs first and can veto the wake before a single token is spent.
- Wake on a GitHub push, a Sentry alert, a Stripe payment, inbound email, a Discord message, or cron
- The guard is your code: it decides, per event, whether this one is worth an agent
Watch every run land on the board
A woken agent appears on your fleet board like any other — its own card, its transcript, its diff, the same review before anything lands. The nightly dependency audit arrives that way, already waiting when you get in.
One run can wake the next
An automation can finish by starting another, so a chain of specialists hands work down the line — triage wakes the fixer, the fixer wakes the reviewer — each a fresh session with its own transcript.