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

# How routines work

> Hand off recurring work: set it up once, and the agent runs it on a schedule.

Your agent knows your stack and has access to your tools. Now teach it to use them on its own.

You probably have recurring work — a Monday performance report, a daily data export, a regular competitor check, a weekly pipeline digest. Routines are how you hand those off: set it up once, and it runs on a schedule.

## Think of it as a standing instruction

A one-off task is "send this email." A Routine is the standing version: "every Monday at 8am, pull last week's numbers and send me a summary." You set the expectation once instead of asking each time.

Because the agent already has your connectors, a Routine is personalized from the start — your weekly report highlights the pages, accounts, and metrics that matter to you, not a generic dump.

## The four pieces

Every Routine has the same structure:

* **When it runs** — the trigger: a daily or weekly schedule, or a one-time future run.
* **What it does** — the instructions, written in plain language.
* **What it can access** — the connectors and tools it is allowed to use.
* **How much you trust it** — the mode (read-only, approval-required, or autonomous).

## A dial, not a switch

Ottic gives you a dial for trust, not an on/off switch. You control how much autonomy the agent has, and you can change it anytime:

* **Read-only** — it can analyze and report, but cannot change anything.
* **Approval-required** — it does the work, then pauses and shows you the output before anything happens.
* **Autonomous** — it just runs and delivers, no waiting.

Most people start with approval-required, review a few runs in the history, then promote to autonomous once they trust the behavior. You can always dial back.

## Build one in conversation

You do not fill out a form. Just describe what you want:

> Every Monday at 8am, pull last week's Search Console and GA4 data for example.com, summarize wins and drops, flag pages that lost rankings, and save it as a shareable report.

The agent assembles the Routine — trigger, instructions, tools, and a suggested mode — and you adjust anything. The more specific you are, the better the first run.

## Every run is logged

Each Routine keeps a full run history — when it ran, what it did, whether it succeeded, and the resulting session and artifacts. That is your audit trail and your debugging tool: instead of guessing why a run looked off, you can open it and see exactly what happened.

## What's next

You know what Routines are and how the trust dial works. Next: the patterns that make the biggest difference — writing better instructions and what power users do differently.


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