clawdbot — the fleet orchestrator
The brain of the operation. clawdbot routes every inbound channel to an isolated agent, owns the cron engine that fires the recurring jobs, and gives every agent the same 52-skill toolbelt.
clawdbot is the difference between “a bunch of scripts” and “a fleet.”
It is an always-on orchestrator that takes every message coming in from Slack, WhatsApp, Discord, iMessage
and 24 other channels, routes it to the right isolated agent (its own workspace, its own memory,
its own session), and hands that agent a shared toolbelt of 52 skills.
What it does
Three jobs, one service: route, schedule, and equip. It routes inbound conversations to
per-agent workspaces, it schedules recurring work through a built-in cron + heartbeat engine, and it equips
every agent with the same skills and channel connectors so behavior is consistent across the fleet.
How the scheduling works
- Cron jobs live in
~/.clawdbot/cron/jobs.json. Each job carries a schedule
(ata timestamp,everyN minutes, or a fullcronexpression with a timezone),
a target session (mainor anisolatedagent), a payload (a system event or a full agent turn,
with an optional model + thinking-level override), and a run-log of its last status. - Heartbeats handle the fuzzy stuff — “check email + calendar + mentions a few times a day” —
batching several loose checks into one wake-up instead of a dozen separate crons. - Isolated runs keep a scheduled job’s history out of your main chat, then post a short summary back.
needs its own model; use a heartbeat when several small checks can ride together. Batching beats a pile of
one-off crons — fewer wake-ups, less token burn.
Why it anchors the recursive system
Every other job on this page is either scheduled by clawdbot or feeds a signal back into it.
That two-way street — it dispatches work, and the results (a cost re-price, an audit, a client request) change what it
does next — is what makes the fleet a loop instead of a to-do list.