Schedules
Every cadence, how it interacts with your plan's data refresh and credits, and how windows and cooldowns fit in.
A schedule says when an agent or rule runs. Meerkats offers a fixed set of cadences in the app, from every five minutes to monthly, plus event-based triggers and a one-off run. Over the API a schedule is a five-field cron expression. This page lists them, explains how they interact with your plan’s data refresh and with credits, and shows how to pick one.
The cadences#
| Cadence | Cron | Typical use | Plan |
|---|---|---|---|
| Event-based | When something happens | A sync completes, an order lands, a watch diverges, a proposal is decided | — |
| Every 5 minutes | */5 * * * * | Spend caps, outage checks. 288 runs a day | $399 (15-minute refresh) for fresh data |
| Every hour | 0 * * * * | Daily-triage monitors on paid plans | $49 and up |
| Every hour (working hours) | 0 9-18 * * 1-5 | Hourly, weekdays, office hours | $49 and up |
| Every hour (8am–11pm) | 0 8-23 * * * | Hourly through the trading day | $49 and up |
| Every day at 10am | 0 10 * * * | Morning reports, daily checks | Any |
| Every weekday at 10am | 0 10 * * 1-5 | The same, skipping weekends | Any |
| Every Monday at 10am | 0 10 * * 1 | Weekly optimisation and reports | Any |
| Monthly (1st at 10am) | 0 10 1 * * | Monthly reviews | Any |
| Once | — | Run now, then stop | Any |
Times are in the workspace time zone. A new rule defaults to every 30 minutes (*/30 * * * *) with a 24-hour cooldown.
Schedules and data refresh#
A rule can only see new data as often as the data refreshes, and that follows your plan:
| Plan | Data refreshed | Fastest useful cadence |
|---|---|---|
| Free | Daily | Every day at 10am |
| $49 | Hourly | Every hour |
| $149 | Every 30 minutes | Every 30 minutes |
| $399 | Every 15 minutes | Every 5 minutes, for live reads |
A five-minute rule on an hourly plan evaluates the same numbers twelve times, using twelve runs’ worth of credits for one hour’s worth of information. Match the cadence to the refresh, and move up a plan when you need to react faster.
Event-based beats polling
For “as soon as it happens”, use an event-based trigger on the sync, not a five-minute poll. The rule runs once per refresh, exactly when there is something new to see.Schedules and credits#
Every scheduled run uses one credit, whether or not the condition is met. Rough monthly usage per rule:
| Cadence | Runs a month | Per workspace |
|---|---|---|
| Every 5 minutes | ≈ 8,640 | Only for a handful of critical checks |
| Every hour | ≈ 720 | Comfortable on Starter for a few rules |
| Every hour (working hours) | ≈ 220 | Hourly at a third of the cost |
| Daily | 30 | Almost free |
| Weekly | 4 or 5 | Almost free |
The counters on Plans & credits show the running total. See Credits and billing.
Cadence tiers#
The cadence you pick also sets what the agent is expected to do on its own. Agents run history groups runs by these tiers.
| Tier | Job | Autonomy |
|---|---|---|
| Daily (and faster) | Triage and protect: fast signals, mostly reversible changes | Can be allowed to act alone within caps once it has a record |
| Weekly | Optimise: enough data for significance | Recommend, then a person approves |
| Monthly | Strategise: structural, slow-signal decisions | The agent frames the options; a person decides |
Windows, cooldowns and schedules together#
Three settings shape when a rule fires, and they are easy to confuse.
| Setting | What it controls | Default |
|---|---|---|
| Schedule | How often the rule looks | Every 30 minutes |
| Window | How many days the condition must hold before it counts | 4 days |
| Cooldown | How long after firing before it can fire again on the same target | 24 hours |
A rule that looks every 30 minutes, with a four-day window and a 24-hour cooldown, fires at most once a day per target, and only on a problem that has lasted four days. Shorten the window for faster reaction; lengthen the cooldown for calmer behaviour.
Platform attribution lag#
Conversions arrive for days after a click. A rule on a conversion metric with a one-day window sees yesterday’s spend with few of yesterday’s conversions, and fires on lag. Meerkats treats spend with zero conversions younger than three days as pending, not failed, and the measurement checks flag an unripe window. For ROAS, CAC and conversion rules, a window of three days or more is safer.
Over the API#
{
"schedule_cron": "0 9-18 * * 1-5", // every hour, weekdays 9 to 18
"cooldown_hours": 24,
"window_days": 3
}Cron fields are minute, hour, day of month, month, day of week, in the workspace time zone.
Troubleshooting#
The rule runs but sees no new data
Its cadence is faster than your plan’s refresh. Check the sync chip in Data Hub › Bronze Layer; slow the rule down or move up a plan.
It fired at a strange hour
Schedules use the workspace time zone, set on the workspace. Check it under Workspaces (pencil icon).
It's eating credits
Five-minute rules use 8,640 runs a month. Change to hourly, or to working hours, unless it really needs to be that fast.