> ## Documentation Index > Fetch the complete documentation index at: https://docs.pinata.cloud/llms.txt > Use this file to discover all available pages before exploring further. # Tasks > Run prompts on a schedule Tasks are scheduled prompts. The agent runs them automatically — every Monday morning, every six hours, once next Tuesday at 2pm, whatever you set. Use them for daily check-ins, periodic reports, automation triggered by time rather than by a message. Image ## Create a task Open your agent → **Tasks** → **+ ADD**. The dialog asks for: 1. **Name** — anything readable, e.g. `daily-check-in` 2. **Schedule** — when it runs 3. **Prompt or system event** — what gets delivered to the agent 4. **Delivery (optional)** — where the response goes if you want it to leave the agent That's it. Save and the task starts running on the schedule. ## Schedule types You have three choices for `schedule.kind`: * **Cron** — standard cron expression. Most flexible. ```text theme={null} 0 9 * * * # Every day at 9am 0 */6 * * * # Every six hours 0 9 * * 1-5 # Weekdays at 9am ``` * **Every** — run every N milliseconds. Set `staggerMs` if you want to randomize the start offset so multiple tasks don't fire at the same instant. * **At** — run once at a specific ISO timestamp. You can set a timezone (`tz`) so cron respects DST and local time. Minimum interval across all kinds is 1 minute — anything tighter is rejected. ## What gets delivered Two payload kinds: * **`agentTurn`** (default) — your task fires a chat message. The agent treats it like a regular conversation turn. You can pin a specific `model`, toggle reasoning on or off via `thinking`, and set a `timeoutSeconds`. Good prompts to schedule: > "Check my portfolio and summarize anything that moved more than 5% today." > > "Generate the weekly status report from the work log." * **`systemEvent`** — fires a system event instead. The agent sees it as a non-conversational trigger. Useful for heartbeats, periodic maintenance, or anything you don't want polluting the chat history. ## Where the response goes By default, the response lands back in the agent's main conversation. If you want it to go somewhere else: * **Announce to a channel** — send the response to Telegram, Slack, or Discord. Set `delivery.mode` to `announce`, plus a channel and recipient. Good for alerts or daily standups. ```json theme={null} "delivery": { "mode": "announce", "channel": "telegram", "to": "123456789" } ``` * **Webhook** — POST it to a URL. Good for hooking into other services. ```json theme={null} "delivery": { "mode": "webhook", "to": "https://example.com/hook" } ``` Set `bestEffort: true` in delivery if you'd rather swallow a delivery failure than mark the whole run as failed. ## Sessions: main vs isolated A task can run in the agent's main conversation (so its output is part of the running chat history) or in an isolated session of its own. Pick one with `sessionTarget`: * `main` — output joins the live conversation * `isolated` — one-off session that doesn't touch the main thread. Good for batch jobs and webhook-style work. ## Wake mode `wakeMode` controls timing: * `now` (default) — fire the task at the scheduled time, even if the agent is busy * `next-heartbeat` — wait for the next heartbeat tick. Useful for low-priority background work. ## Managing tasks Each task in the list has: * **Run now** — fire it immediately, doesn't affect the schedule * **Enable / disable** — pause without deleting * **Edit** — change schedule, prompt, or delivery * **Delete** — remove it * **Runs** — open the run history Run history shows what fired and when, and whether each run succeeded. ## From the CLI ```bash theme={null} pinata agents tasks list pinata agents tasks create pinata agents tasks run pinata agents tasks delete ``` ## Limits and gotchas * **20 tasks** per agent * Minimum interval **1 minute** * A task that runs longer than `timeoutSeconds` (or 5 minutes if unset) is cancelled and marked failed * Tasks survive agent restarts. Schedules that would have fired during downtime are skipped, not queued.