Schedules

Run your agent automatically on a timer — daily digests, weekly reports, one-off tasks, or any cron schedule.

Schedules let your agent run itself without anyone sending a message. Set a time, describe what you want done, and the agent fires automatically on that cadence.

Common uses: sending a daily summary to a Slack channel, generating a weekly report, checking an external feed and emailing you if something changed, running a one-off cleanup task at a specific date and time.

Setting up a schedule

Go to your agent’s Schedules page in the dashboard and click Add schedule.

Each schedule has two parts:

1. When to run — pick a frequency and time, or write a raw cron expression
2. Task notes — the instruction the agent receives each time it fires (what to actually do)

Frequency options

Option	Description
Once	Runs one time on a specific date and time, then stops
Daily	Runs every day at the time you choose
Weekly	Runs on a specific day of the week
Monthly	Runs on a specific day of the month
Custom (cron)	Full 5-field cron expression for any schedule

All times are stored in the timezone you choose. UTC is the default.

Cron expressions

Switch a schedule to Cron mode to enter a standard 5-field expression:

┌───── minute (0–59)
│ ┌───── hour (0–23)
│ │ ┌───── day of month (1–31)
│ │ │ ┌───── month (1–12)
│ │ │ │ ┌───── day of week (0–6, 0 = Sunday)
│ │ │ │ │
* * * * *

Examples:

Expression	Meaning
0 9 * * 1	Every Monday at 09:00
30 17 * * 1-5	Weekdays at 17:30
0 8 1 * *	First of every month at 08:00
*/15 * * * *	Every 15 minutes

Task notes

Task notes are the instruction the agent sees when the schedule fires — think of it as the message you’d type if you were triggering it manually.

Be specific. Instead of “do the weekly report”, write something like:

Pull this week’s activity summary and post it to the #updates Slack channel. Flag any items where revenue dropped more than 10% week-over-week.

The agent runs with its full configuration — its purpose, tools, connectors, and memory — so you only need to describe the specific task for this run, not re-explain everything.

Delivery

By default a scheduled run’s output is saved to the schedule’s History and visible in the History modal — that’s it. To push the output somewhere, pick a Delivery target on the schedule:

Delivery kind	Target
No delivery	(default) — result shown in the History modal only.
Email	An email address.
Slack	A channel id (e.g. C01234567) or #channel-name.
Telegram	A numeric chat id.
Webhook (HTTPS POST)	An https://… URL. The agent POSTs the output to it.
Inbox (in-app conversation)	Leaves the result as a conversation in the agent’s Inbox.

You also pick a Delivery format — markdown (default), text, json, or artifacts_only.

Run guards

Each schedule has three per-run guards you can tune:

• Max steps — maximum LLM/tool steps for a single run (1–100, default 20).
• Timeout — maximum wall-clock duration per run, 5s–10min (default 2 minutes).
• Credit budget — optional ceiling on credits a single run is allowed to spend (1–1,000,000). Leave blank for no limit.

The runtime aborts the run as soon as any guard trips.

Spend budget

Separate from the per-run Credit budget guard, each schedule has a cumulative Spend budget — a ceiling on the total credits this schedule may spend across a period, not just within a single run. It lives in its own fieldset on the schedule card:

• Spend cap (credits) — the ceiling (up to 1,000,000). Leave 0 for no limit.
• Budget period — Daily (per day), Weekly (Mon–Sun), or Monthly (calendar month).

When a cap is set, the card shows a live Spent this period: X / Y credits tracker. Once cumulative spend reaches the cap, every further run is skipped — the card shows budget reached — runs skipped until the period resets — until the period rolls over (UTC), when the counter resets to zero. This is independent of the per-run credit budget: the per-run guard caps one run, the spend budget caps the schedule’s total over the period.

Examples

Hello world — a daily reminder. A schedule that fires each morning with task notes “Post ‘Good morning! Standup is at 9:30.’ to the team.” Delivery: Slack → #team. The simplest autonomous run.

Simple — a weekly digest. Fire every Monday at 8am. Task notes: “Summarise last week’s activity and email it to me.” Delivery: Email. The agent runs with its full config (tools, data, memory) and produces the digest unattended.

Medium — a monitored report with guards. Fire daily. Task notes: “Pull yesterday’s sales numbers via the orders connector, flag anything down >10% week-over-week, and post to #updates.” Set a Credit budget of 200 and a Timeout of 3 minutes so a misbehaving run can’t overspend. Delivery: Slack, format markdown.

Complex — a multi-step autonomous workflow with webhook delivery. Fire hourly during business hours. Task notes describe a multi-step task: “Check the support inbox via the connector, triage any new tickets by severity, draft responses for the low-severity ones, and for high-severity ones gather the relevant account context.” Raise Max steps to 40 for the multi-tool work, set a Credit budget ceiling, and deliver to a Webhook that feeds your ticketing system. Use Run now to test it before trusting it on a schedule, and watch the History modal to review each run’s detailed execution trace and tune the task notes until the runs are reliable.

Complex — a hard monthly spend ceiling. Cap total spend, not just per-run: set Spend budget → cap 2,000 credits, period Monthly. An hourly schedule can burn through its per-run 50-credit cap on every fire, but once cumulative spend for the month hits 2,000 the card shows budget reached and every further run is skipped until the 1st resets it — a hard monthly ceiling independent of the per-run cap.

Run now and history

Each schedule card has a Run now button — fires the schedule immediately as a manual run. The result shows up in the History modal alongside automated runs.

Click History on a card to see the most recent 20 runs for that schedule, including trigger (manual vs cron), status, duration, credits spent, delivery outcome, and a preview of the output. Each run links to a detailed execution trace.

Enable and disable

Each schedule has an ON / OFF toggle. Disabled schedules are preserved with their configuration but don’t fire. You can re-enable them at any time without reconfiguring.

Activity log

Every scheduled run shows up in the Activity page the same as any other conversation. You’ll see the task notes that triggered the run, the agent’s full response, tool usage, and cost.

Limits

• Up to 50 active schedules per agent (this limit may change).
• Task notes: up to 4,000 characters.
• Minimum interval: 1 minute (for cron mode).
• Schedules are owner-scoped — for subscriber scheduling, see the subscriber dashboard’s own Schedules page.

Manage via the Management MCP

This area is managed through the Management MCP at https://manage.agentheya.com/mcp.

Endpoint	https://manage.agentheya.com/mcp
Tools	list_schedules, create_schedule, update_schedule, delete_schedule

All tools take owner_id + agent_id as required arguments. See the Management MCP guide for authentication and the full tool catalogue.