Meta’s Manus Brings Its AI Agent to the Desktop With My Computer Feature

San Francisco: Manus, the AI agent startup acquired by Meta in late 2025 for approximately $2 billion, has launched a desktop application for macOS and Windows. Its core feature is called My Computer. It gives the Manus agent direct access to a user’s local files, applications, and terminal, tasks that previously required uploading data to a remote server.

The shift is significant. Until now, Manus operated entirely through a web interface. Users gave it a task, it ran in a cloud sandbox, and returned results. My Computer collapses that gap. The agent can now read, edit, and organise local files, execute command-line instructions, launch applications, and tap into a machine’s own GPU for inference tasks.

Similar to Perplexity’s Personal Computer, which automates tasks on Mac, Manus brings AI-driven productivity to the desktop by integrating with local systems, allowing users to streamline workflows directly on their machines.

Also read: Picsart Launches AI Agent Marketplace to Automate Creator Workflows

Practical use cases include sorting thousands of unsorted photos into labelled folders, renaming batches of invoices, and building applications using programming environments already installed on the machine. In one demonstration, Manus built a real-time translation and subtitle app on macOS entirely through terminal commands, no manual coding required.

Also read: WordPress.com Enables AI Agents Write, Edit, Publish Content

The launch is a direct response to OpenClaw, the open-source desktop AI agent that went viral earlier this year. Jensen Huang called OpenClaw the “next ChatGPT.”

Its creator Peter Steinberger has since joined OpenAI. Unlike OpenClaw, which is free and open-source, Manus runs on Meta’s proprietary model stack as a paid subscription. This model of charging mirrors the wider trend, where AI agents are pushing companies to reconsider the traditional per-seat pricing structure.

To address privacy concerns, all terminal commands require explicit user approval before execution. Users can approve actions individually or set trusted recurring tasks to run automatically.