Skip to main content
It’s Friday, 6 PM. The office is empty. On its own schedule, Athena pulls the week’s activity, compiles the summary, drops it in the shared folder, and pings the owners — the report that used to eat your Friday evening is already waiting when you log off. That’s a routine: the work that should happen on a clock, happening without anyone remembering to press go. It’s how your company keeps running while no one is watching the keyboard.

What routines are for

Routines are tasks that repeat on a fixed cadence — every Monday at 9 AM, every weekday at 6 PM, every first day of the month. Instead of you remembering to do it + executing it manually, a routine fires automatically at the configured time and runs what you defined — which can involve an agent, creating a task, sending an outbound, or a combination.

When to use a routine vs a task

ScenarioUse
”I need to do X once, next week”Task with a date
”I need to do X every week, indefinitely”Routine
”I want the agent to review stalled leads every Tuesday”Routine + agent
”I’m going to launch a campaign now for new leads”Outbound (one-shot)
“I want a weekly email report every Friday at 6 PM”Routine + Marcus
The general rule: if you’ll need to do it again next week, it’s worth creating a routine.

Anatomy of a routine

Cadence

When it fires — daily / weekly / monthly / custom cron expression. Supports the org’s timezone.

Action

What runs when it fires — call an agent, create a task, send a message, run an outbound pipeline.

Context

Inputs the routine passes to the action (e.g., “leads with no activity in the last 7 days”). Can be static or dynamic.

History

Every execution is recorded — when it ran, what it did, whether it succeeded, and how many Stars it spent.

Common patterns

Cadence: every Friday at 6 PM Action: Athena reviews the week’s activity and compiles a summary document Output: new doc in a shared folder + chat notification to ownersUseful for: recurring visibility without manual effort.
Cadence: every Tuesday at 10 AM Action: Marcus reviews leads with no activity in the last 14 days and proposes follow-up drafts (HITL on — you approve) Output: drafts waiting for approval in chatUseful for: keeping opportunities from dying due to lack of follow-through.
Cadence: first day of the month at 8 AM Action: Athena archives leads in terminal state for more than 90 days + identifies duplicates Output: review task for the pipeline ownerUseful for: preventing the pipeline from getting inflated with historical noise.
Cadence: every weekday at 7 AM Action: runs an Apify actor (Maps lead search with a pre-defined query) and adds new results to the “New Outbound” pipeline Output: N new leads ready for Marcus to workUseful for: keeping the funnel fed without manual effort.

Setting up a routine

1

Sidebar → Routines → New routine

Opens the configurator.Expected result: panel with 4 tabs (Cadence / Action / Context / Notification).
2

Define the cadence

Options:
  • Daily (choose a time)
  • Weekly (days of the week + time)
  • Monthly (day of the month + time)
  • Custom (cron expression — for complex scenarios)
For cron expressions, standard Linux format: 0 18 * * 5 = Friday at 6 PM. See crontab.guru to test your expression.
3

Choose the action

What will run:
  • Call agent — passes a pre-defined prompt to Scout / Marcus / Athena
  • Create task — new task on a specific board
  • Trigger outbound — run an existing campaign on a filtered audience
  • Execute tool — e.g., run maps_lead_search directly
Each type opens its own configurator.
4

Define notification (optional)

When the routine finishes, who gets notified:
  • Nobody (silent)
  • You (inbox notification)
  • Specific people (mention)
  • External channel (Slack via Composio)
5

Activate

Toggle Active. The next scheduled execution appears below — confirm the cadence looks right before closing.

Limits and gotchas

Routines consume Stars — just like any billable action. A routine that calls an agent daily will debit Stars every day. Review the cost estimate before activating; it appears in the configurator based on the org’s history.

Agent daily cap applies

If a routine triggers an agent that has a daily cap, and the routine exceeds that cap, the execution is blocked. You’re notified to decide: raise the cap or accept the one-off failure.

Skip during sensitive windows

Org settings can define sensitive windows (weekends, holidays, after-hours) where routines don’t fire outbound to customers. Internal-only routines keep running.

Idempotency per execution

If a routine fails and retries, it uses an Idempotency-Key to prevent double-firing. You won’t accidentally send the same message twice.

Backfill off by default

If a routine was paused for 3 days and you reactivate it, it does not run the 3 missed executions. It only runs the next scheduled one. For an explicit backfill, use “Run now” manually.

Audit log

Every routine execution creates an entry in the history:
FieldWhat it records
executed_atExecution timestamp
routine_id + routine_nameWhich routine
action_typeAction type (agent / task / outbound / tool)
statussuccess / failed / skipped
cost_starsStars spent
output_refLink to the created resource (task, draft, etc.)
Owners and Admins can inspect this under Routines → click a routine → History tab.

Next steps

Outbound

How Marcus operates the campaigns that routines can trigger.

Agents

Who you can call from within a routine.

Caps

How to limit spending from routines that run out of control.

Composio

For external integrations that routines trigger (Slack notifications, Calendar, Notion).