Reach out

Cron

Brigade can do work on a schedule even when no one is watching. Jobs run inside the gateway, survive restarts, and can fire on a cron expression, a fixed interval, or a single future timestamp.

Three ways to schedule#

Pick exactly one of these per job:

• --cron "0 9 * * *" — a cron expression. Pair it with an IANA --tz.
• --every — a fixed interval.
• --at "2026-06-04T09:00:00+05:30" — a one-shot timestamp.

$ brigade cron add --cron "0 9 * * *" --tz "Asia/Kolkata" --message "good morning"$ brigade cron add --at "2026-06-04T09:00:00+05:30" --message "stand-up reminder"

What a job does#

A job's payload can be any of:

• A system event into your main session (--message) — like a nudge you would have typed.
• A full agent turn in an isolated session (--model, --tools, --light-context).
• An optional delivery target (--channel, --to) so the result lands in a channel.

Cron supports stagger so a burst of jobs does not all fire on the same tick, and each job records a per-run log you can inspect.

Managing jobs#

$ brigade cron list$ brigade cron status$ brigade cron runs <jobId>        # per-job run history$ brigade cron run <jobId>         # fire it now$ brigade cron enable <jobId>      # / disable / rm <jobId>$ brigade cron edit <jobId>

Heartbeat#

Cron fires discrete jobs; the heartbeat is the other half of always-on. Driven by a workspace HEARTBEAT.md, it wakes an agent on a regular interval to drain any queued events and do periodic work. Each wake is gated: it honours a global enable flag, yields if your main turn or the session is already busy, and — except on a plain interval tick — skips when there is nothing pending. An interval wake proceeds even with an empty inbox, so the agent can do scheduled work without an incoming message.