Manual
Run when you press a button
Schedule
Run on a time-based schedule
On Complete
Run after another task finishes
Integrations
Run when something happens in a connected app
Manual
The simplest trigger. The task runs when you press the Run now button — nothing fires automatically.Each agent can have at most one manual task. This is the “on demand” button you always have available.
- Deep-dive research you want to kick off yourself
- Reports you generate occasionally, not on a fixed schedule
- Testing an agent before automating it further
- One-off tasks where timing doesn’t matter
- “Analyze this week’s data” — you run it whenever you’re ready to review
- “Draft a proposal for [client]” — triggered when you start working on a pitch
- “Summarize my notes” — run before a meeting when you need a quick brief
Schedule
The task runs automatically at a set time or interval. No manual action needed. Frequency options:| Frequency | Example |
|---|---|
| Once | Fire at a specific date and time |
| Minute | Every N minutes |
| Hourly | Every hour, or every N hours |
| Daily | Every day at a set time |
| Weekly | On specific days of the week |
| Monthly | On a specific date or day (e.g., 1st Monday) |
| Yearly | Once a year |
| Custom | Every N minutes / hours / days / weeks / months |
- Time — pick the time in 15-minute intervals
- Timezone — your local timezone is pre-selected; change it if needed
- Start date — optional; the task won’t fire before this date
- End condition — run forever, stop after N runs, or stop on a specific date
- Daily digest — daily at 8am
- Weekly report — every Monday at 9am
- Metrics check — every hour
- Monthly invoice summary — 1st of each month at 7am
On Complete
The task fires when another task finishes. This is how you build task chains — multi-step agent pipelines where each step hands off to the next.- Source task — which task to watch
- Fire when — Succeeds / Fails / Either
- Pass upstream output — inject the previous task’s response into this task’s context
- Building a research → edit → publish pipeline
- Handling errors: run a fallback agent when the main task fails
- Sequential steps where each stage needs the previous one’s output
- Researcher → Editor → Publisher — three agents, each improving the previous output
- Monitor → Alert — a watcher task fails, fires an incident-reporting agent
- Fetch → Format → Send — gather data, format it, post to Slack
Chains are limited to 10 steps. One task can fan out to at most 5 downstream tasks. Cycles (A → B → A) are blocked at
save time.
Integrations
Trigger a task when something happens in a connected app — a new issue in Linear, a payment in Stripe, a webhook from any connected tool. Available integrations:| Integration | Example triggers |
|---|---|
| Linear | Issue created, status changed, comment added, assigned |
| Stripe | Payment succeeded, subscription cancelled, checkout completed |
| Connected apps | Any event from apps connected via your workspace integrations |
Connect an integration from your workspace settings. Once connected, its trigger events appear in the trigger picker.
- New Linear issue → agent triages it, adds labels, and posts a summary to #engineering
- Payment failed → agent drafts a customer recovery email
- Subscription cancelled → agent logs context and notifies the team in #revenue
Choosing the right trigger
| You want to… | Use |
|---|---|
| Run the agent whenever you feel like it | Manual |
| Run automatically every morning | Schedule — daily |
| Run on a repeating cycle | Schedule — hourly / weekly / monthly |
| React to another task finishing | On Complete |
| React to an external event (Linear issue, Stripe payment) | Integration trigger |
| Test an agent before automating it | Manual first, then switch |
Back to Tasks
Learn how tasks connect to agents and how to set up your first automation