With ‘Cloud Computer' Use, Manus Wants Users To Have Agents That Run 24/7 In the Background

In the intensifying competition among AI labs, dozens of new large language models have been released in recent months.

Across China and the U.S., the focus has largely remained on raw model performance and frequent updates to maintain visibility. Yet amid this emphasis on ever-larger or more specialized models, one system has taken a different path. Manus, now part of Meta, operates as a general AI agent designed to move beyond generating text or answers and instead execute practical tasks in the real world.

Its latest addition, the 'Cloud Computer,' highlights this distinction by providing a persistent, always-on environment that addresses a core limitation of most AI tools: the inability to sustain work once a session ends or a device powers down.

According to the announcement on its website, Cloud Computer functions as a dedicated virtual machine in the cloud, running Ubuntu and managed entirely through natural language instructions.

Unlike a standard temporary sandbox session which spins up for a single task and then disappears, the Cloud Computer remains active continuously.

Files, databases, installed tools, and configurations persist across interactions, allowing the agent to build on previous work without starting from scratch each time. This setup eliminates the need for users to configure servers, manage credentials, or write code themselves. A user simply describes the desired outcome, and Manus handles the implementation, from installing software to scheduling operations.

This persistence opens up practical applications that go beyond one-off queries. For instance, it supports 24/7 bots on platforms like Slack, Discord, or Telegram that continue responding to messages or performing routine duties indefinitely.

It can maintain live databases or knowledge bases that accumulate and reference data over weeks or months, such as updating sales records from uploaded files and generating trend reports on a schedule.

Open-source tools like WordPress, Metabase, Home Assistant, or Odoo can be self-hosted and kept running without manual oversight.

Scheduled tasks, including web scrapers that check for updates at specific times and compile reports, also become straightforward. Even command-line developer tools, which often require complex setup, can be deployed and operated through plain-English prompts.

Access to the Cloud Computer integrates directly into the Manus interface on both web and mobile. It can be created via the settings menu by selecting a plan based on resource needs, ranging from basic for simple scripts to more advanced options for databases or team use, and specifying location and storage preferences.

Manus automatically routes tasks to it when continuous operation or shared state is required, while defaulting to temporary sandboxes for isolated work.

A separate Manus Desktop option exists for interacting with files and applications on a user's local machine, keeping the cloud environment distinct and isolated from personal devices for security.

In a landscape where AI agents often falter once disconnected from an active chat, the Cloud Computer represents a deliberate step toward making automated workflows reliable and ongoing.

It shifts the experience from episodic assistance to sustained capability, without requiring users to learn infrastructure management or leave devices running overnight.

As the broader AI field continues its rapid model releases, developments like this underscore a parallel track: not just improving what an agent can reason about, but ensuring it can deliver results that endure.