# Run on a schedule

A nightly digest, a weekly check, a reminder next Tuesday at 9:00 in your time zone.

Start tasks from your own scheduler, or have an owner or admin create an
**automation** on the dashboard. For the agent to pick its own times, use
[follow-ups](/patterns/agent-follow-ups).

**You need:** a deployed runtime ([Deploy](/ship)).

## 1. Test the prompt on staging

```bash
introspection tasks create --runtime digest-agent --environment staging \
  --prompt "Summarize yesterday's tickets and write /workspace/outputs/digest.md"
```

Read the result in [Conversations](/observe/conversations) and adjust the prompt.

## 2. Use your own scheduler

Run the command from cron, CI, or your cloud scheduler with a project API key
([set the variables](/patterns/trigger#1-give-your-service-the-key)):

```bash
0 6 * * * introspection tasks create --runtime-id <runtime-id> --environment production --prompt "Summarize yesterday's tickets"
```

To keep history and memory across runs, post into one task instead:

```bash
0 9 * * 1 introspection tasks prompt <task-id> --prompt "Check what changed since last week and update the summary."
```

A key refuses `--runtime`: pass any version ID of the runtime
(`introspection runtimes list --runtime digest-agent`); the task runs the version
production currently serves. Most schedulers run in UTC.

## 3. Schedule on the platform (owners and admins)

Instead of step 2, use the dashboard's **Automations** page. It appears only when Introspection
has turned Operator on for your organization (contact support). The CLI login
and API keys lack the automations scopes, so the CLI's `automations` commands
return 403.

| Field | Meaning |
| --- | --- |
| Runtime group | Its production version runs the firing. |
| Cron and time zone | Repeat, in an IANA zone; UTC when omitted. |
| Run once | A future time with an offset, such as `2026-11-03T09:00:00+00:00`. |
| Task | Each firing is a new turn in that task, so the agent sees earlier runs. |

**Set a task** unless the agent reports to Slack. An automation without a task
starts a new task as the platform, and is built for check-ins that report to
your organization's Slack integration: without a live one it is skipped with
"Slack is not configured". A Slack thread's task replies in that thread
([post digests](/patterns/slack#post-digests-and-reminders)).

Owners and admins can target any task. Other members can't use this page. An
end user signed in through an application with the automations scopes can have
the agent schedule check-ins inside their own conversation
([follow-ups](/patterns/agent-follow-ups)), up to 100 per member.

## Verify

Check **Next run** on the Automations page (or your scheduler's log), then:

```bash
introspection tasks list --limit 5 -o table
```

Expect a new task, or a new turn in the target task.

Or ask your agent: "Run digest-agent every weekday at 06:00 London time from our CI scheduler."

## Limits

| Not supported | Do this instead |
| --- | --- |
| Creating automations with the CLI login or an API key | The dashboard's Automations page, or your own scheduler. |
| More than 100 follow-ups per member | Delete stale ones; owners and admins have no cap. |
| A run as one of your users (`user_id`) | Target that user's existing task, or run `tasks create --subject <user-id>` from your scheduler. |

Run on a schedule

A nightly digest, a weekly check, a reminder next Tuesday at 9:00 in your time zone.

Start tasks from your own scheduler, or have an owner or admin create an automation on the dashboard. For the agent to pick its own times, use follow-ups.

You need: a deployed runtime (Deploy).

1. Test the prompt on staging


introspection tasks create --runtime digest-agent --environment staging \
  --prompt "Summarize yesterday's tickets and write /workspace/outputs/digest.md"

Read the result in Conversations and adjust the prompt.

2. Use your own scheduler

Run the command from cron, CI, or your cloud scheduler with a project API key (set the variables):


0 6 * * * introspection tasks create --runtime-id <runtime-id> --environment production --prompt "Summarize yesterday's tickets"

To keep history and memory across runs, post into one task instead:


0 9 * * 1 introspection tasks prompt <task-id> --prompt "Check what changed since last week and update the summary."

A key refuses --runtime: pass any version ID of the runtime (introspection runtimes list --runtime digest-agent); the task runs the version production currently serves. Most schedulers run in UTC.

3. Schedule on the platform (owners and admins)

Instead of step 2, use the dashboard’s Automations page. It appears only when Introspection has turned Operator on for your organization (contact support). The CLI login and API keys lack the automations scopes, so the CLI’s automations commands return 403.

Field	Meaning
Runtime group	Its production version runs the firing.
Cron and time zone	Repeat, in an IANA zone; UTC when omitted.
Run once	A future time with an offset, such as `2026-11-03T09:00:00+00:00`.
Task	Each firing is a new turn in that task, so the agent sees earlier runs.

Set a task unless the agent reports to Slack. An automation without a task starts a new task as the platform, and is built for check-ins that report to your organization’s Slack integration: without a live one it is skipped with “Slack is not configured”. A Slack thread’s task replies in that thread (post digests).

Owners and admins can target any task. Other members can’t use this page. An end user signed in through an application with the automations scopes can have the agent schedule check-ins inside their own conversation (follow-ups), up to 100 per member.

Verify

Check Next run on the Automations page (or your scheduler’s log), then:


introspection tasks list --limit 5 -o table

Expect a new task, or a new turn in the target task.

Or ask your agent: “Run digest-agent every weekday at 06:00 London time from our CI scheduler.”

Limits

Not supported	Do this instead
Creating automations with the CLI login or an API key	The dashboard’s Automations page, or your own scheduler.
More than 100 follow-ups per member	Delete stale ones; owners and admins have no cap.
A run as one of your users (`user_id`)	Target that user’s existing task, or run `tasks create --subject <user-id>` from your scheduler.