> ## Documentation Index
> Fetch the complete documentation index at: https://docs.archetype.ing/llms.txt
> Use this file to discover all available pages before exploring further.

# Automations

> Requests that run on a schedule or when something happens, and what each run leaves behind.

Automations live in **Manage → Automations → Schedules**. The page has two lists: schedules, which run at set times, and triggers, which run when an event comes in. `/schedules` and `/triggers` list them from inside a conversation.

## Schedules

A schedule sends a request on its own, like a morning summary or a weekly check. Press **New schedule**:

| Field | What you set |
| - | - |
| **Name** | Lowercase letters, digits and `-` |
| **What should it do?** | The request |
| **When** | **Once**, every few minutes, on chosen days at a time, or a cron expression. Times are in your time zone |
| **Where** | The conversation where it runs, or a chat to answer in |
| **Only send news (a check, quiet at night)** | Answers only when there's something to say, and never during quiet hours |
| **Importance** | How a failed run is reported (below) |

You can also just ask in a conversation, "remind me at 9 tomorrow", and the agent makes the schedule. If you ask in a chat channel, the answer comes back to that chat.

Schedules run while the app is open. When it's closed, the background service runs them (`archetype daemon install`, see [CLI](https://docs.archetype.ing/en/start/cli.md)). If the computer was off, only the most recent missed run happens, and only once. At most 4 schedules run at the same time; the others wait their turn.

A schedule that runs a command with no model at all can only be made from the terminal: `archetype --create-schedule … -- PROGRAM`.

## Triggers

A trigger runs a request when something happens. Press **New trigger** and choose what wakes it up:

* **An event from a plugin.** Some ready-made ones: **When a payment comes in**, **Before a meeting**, **When mail wants a reply**, **When a todo is due today**, **When someone's birthday or date comes**. Narrow it down by the event's fields. Text matches regardless of case, and a `*` at the start or end matches part of a word. Events that arrive within 20 seconds of each other run together in one turn.
* **When an event is posted**: another service (a failed build, a new issue) calls an address. The address carries a secret token. Turn on **Require a signature (recommended)** to get a secret you paste into a GitHub webhook. It's **Shown only now**. Each event runs only once, even if it's sent again.

The event goes along with the request as data, not as instructions.

## What a run leaves behind

* **Its conversation.** Each run is a turn in the automation's conversation (the one you picked, or a conversation of its own). **Open its conversation** opens it. A turn nobody asked for in person says what started it.
* **Recent runs**: when each run happened and how it ended. **Next run** shows when the next one is due. **Run now** runs it right away.
* **A failed run**, reported according to **Importance**: **Pinned** tells you right away, **Normal** lists it in **Needs you** and tells your devices, **Background** keeps it in the record only.
* **An approval the run asked for.** By default, it waits up to 60 minutes and is then denied, and the run reports what it couldn't do. With **Unattended approvals** on, the run stops and the request waits for you in [Arch](https://docs.archetype.ing/en/use/arch.md). Your answer continues it.
* **Output**: for a command schedule, the output of its last 20 runs.

Removing an automation stops it. Its conversation and answers stay.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.