---
title: "Tasks"
id: "2134"
type: "page"
slug: "operations"
published_at: "2026-10-04T06:34:25+00:00"
modified_at: "2026-10-22T16:38:23+00:00"
url: "https://xedant.com/agents/multiagent/docs/operations/"
markdown_url: "https://xedant.com/agents/multiagent/docs/operations.md"
excerpt: "Long jobs — installing and updating agents, recreating containers, moving an agent to another factory,…"
---

# Tasks

[https://xedant.com/agents/multiagent/docs/operations/.md](https://xedant.com/agents/multiagent/docs/operations/.md)

Long jobs — installing and updating agents, recreating containers, moving an agent to another factory, erasing, bulk checks — take minutes. They used to keep you watching a log and holding the browser open. Now such work goes into the task queue: start it and walk away, the work runs on its own and its log is kept.

## The Queue of Jobs

Jobs of one server run strictly one at a time, in the order they were started. That way heavy operations do not get in each other’s way and do not compete for the same machine resources. Fleet-wide jobs — stopping or restarting all agents, freezing the fleet, rolling settings back — run on their own lane and do not wait for routine maintenance: in an emergency the command runs right away.

Starting a task does not open a panel or take you away from the current page: the counter **“Tasks (N)”** grows in the sidebar footer and a short notice appears. The task panel opens with the same button and lives right where you talk to your own assistant — it is the second mode of the shared panel, and chats and tasks switch in it in place. More on the [Own Agent](/agents/multiagent/docs/own-agent)
 page.

## The Task Log

While the work runs, its log updates in real time: every line appears as it is executed. When the work finishes the log stays — you can come back later and see how yesterday’s update ended. Logs are kept for a week, and the list holds up to 2000 recent jobs.

Finished work is not resumed: if MultiAgent restarted while a task was running, the task is marked **“Interrupted”** and has to be started again. So you are never left wondering whether some job broke off halfway.

## Stopping a Job

Any active task can be stopped with the **“Stop task”** button on its row. The state changes to **“Cancelled”** immediately — the panel does not wait for the work to actually finish. The work itself breaks off at the nearest safe step: a single operation at the first opportunity, a batch one after the current row. Everything that arrived before the stop stays in the log.

## The Task List

There is a single list, with no switches or filters: fresh jobs on top, and the order changes by itself when a task starts. A row shows when it began (for example “5 minutes ago”), how long it lasts — waiting, working or total time — and a state badge: **“Queued”**, **“Running”**, **“Completed”**, **“Error”** or **“Cancelled”**. The **“Clear completed”** button removes finished rows and leaves only active jobs.

## What Stays Instant

Not everything goes into the queue. **“Check and fix”**, **“Check connection”**, the server terminal and file copying to a server run right away and show their result in place — they are short and need you to be there. The fixes that “Check and fix” starts already run through the queue. The **“Logs”** tabs of agents and tools show container logs, not tasks, and work as before.

[← Back to documentation](/agents/multiagent/docs)