---
title: Workers
sidebar_title: Overview
description: Workers run on their own — on a schedule or when you ask — read what they're allowed to, do the job, and stop for your approval before anything leaves your account.
---

A **worker** is work that runs without you starting it. On a schedule or on
demand, it reads the sources it was given, does its job, and reports back.
Anything it wants to send or post stops for your approval first.

![The Workers page empty state: "A worker runs on its own — on a schedule or when you ask — and reports back. It can read your Slack and Jira context, search the web, work a mailbox, or watch a URL. Anything it wants to send or post stops for your approval first."](../assets/screens/workers-empty.webp){ width="1600" height="510" }

## What workers do well

- **A Monday brief** — summarise last week's incidents from Slack and Jira and
  prepare an update for your approval.
- **A support digest** — read the support mailbox, group what customers hit,
  draft replies for review.
- **Market or competitor watch** — research a topic on the web each week and
  report sources.
- **Uptime monitoring** — check a public URL every few minutes and report
  outages and recoveries.

## Kinds of worker

| Kind | What it does | Needs |
|---|---|---|
| **General** | Follows its instructions with only the tools and sources you grant: company context, web, mailbox, Slack proposals. | Grants for whatever it uses. |
| **Research** | Runs a bounded public web research query and reports results with sources. | Web search. |
| **Mail** | Reads a mailbox, drafts replies, and sends only exact drafts you approve. | Mail read (+ draft, + send). |
| **Uptime monitor** | Checks a public URL on a schedule and records outages and recoveries. | HTTP read. |

See [Worker types](types.md) for each in detail.

## How a worker is put together

- **Agent** — the worker's identity and instructions, and the capabilities it
  may ever use.
- **Worker** — one deployment of an agent: its kind, trigger, grants, sources,
  credentials, approval rules, budget and default input.
- **Version** — every edit creates a new version. A run always uses the exact
  version it started with.
- **Run** — one execution, with its steps, result and any approvals.

## Lifecycle

```mermaid
stateDiagram-v2
    direction LR
    [*] --> Draft
    Draft --> Active: activate
    Active --> Paused: pause
    Paused --> Active: resume
    Active --> Archived
    Paused --> Archived
    Draft --> Archived
```

Only **active** workers run on schedule. Any worker can be **run now** for a
test.

<div class="phones" markdown>

![Workers on mobile: "Support digest" waiting to send an email to a customer with Send it and Discard; "Release announcer" waiting to post to Slack with Approve and No; and recent runs, one failed because a Gmail credential expired](../assets/screens/mobile/workers.webp){ width="585" height="1266" }
/// caption
The phone is the approval queue for workers.
///

</div>

## In this section

<div class="cards" markdown>

- [**Create a worker**](create.md)

    Describe it in plain language, then review grants, sources and approvals.

- [**Schedules and triggers**](schedules.md)

    Manual, cron with a time zone, or a fixed interval.

- [**Runs and approvals**](runs.md)

    What a run goes through, and how exact-content approvals work.

- [**Worker types**](types.md)

    General, research, mail and uptime workers in detail.

</div>