Automations
Run an agent on a schedule, or loop it against a goal until your checks pass — without starting each thread by hand.
Automations run an agent without you starting the thread. Open them from the sidebar. There are two kinds, on their own tabs.
Schedules
A schedule runs one prompt, on one agent, in one working directory, on a cadence — either a fixed interval or a cron expression with a time zone.
Each schedule chooses where its run lands: a new thread each run, or injected into an existing thread so the work accumulates in one place. It also decides what happens when the app wasn't running at the appointed time — follow the default, catch up once, or skip the missed window entirely.
A schedule can be paused and resumed, or triggered immediately with Run now. Its run history lists every run as Running, Succeeded, Failed, or Skipped, and each one opens the thread it produced.
Loops
A loop is for work you can verify mechanically: give it a goal and the checks that prove the goal is met, and it keeps trying until they pass.
Each round runs the goal in a fresh worker thread, then runs your verify checks in order — shell commands that must all exit 0. You can additionally name a verifier agent and describe in words what counts as done, for goals a command can't judge on its own. At least one of the two is required.
You set a maximum number of iterations and an optional delay between them; the loop stops when verification passes, when the cap is reached, or when you stop it. Each iteration shows its checks, the verdict, and its log, and links to the worker and verifier threads so you can read what actually happened.
Loops run agents unattended
Every iteration is a real agent run against your working directory, with whatever approval policy that agent has. Point a loop at a checkout you're comfortable letting an agent change repeatedly.