Mark Zuckerberg’s agent strategy is taking shape: following the acquisition of Manus, a Chinese AI-agent startup, and the takeover of Moltbook, the new subsidiary is bringing something to market that bears a striking resemblance to OpenClaw.
Meta’s subsidiary Manus has unveiled a new feature that brings AI agents directly to users’ local computers. The feature, called “My Computer,” is part of the new Manus desktop application and is now available for macOS and Windows. Observers are drawing parallels to OpenClaw, the personal AI assistant created by Peter Steinberger, who recently joined OpenAI (and whom Zuckerberg had actually wanted on his team).
Most recently, Perplexity also launched a very similar application called “Personal Computer,” and in China in particular, tech companies have already fully jumped on the OpenClaw bandwagon (more on that here).
What “My Computer” is supposed to do
Until now, Manus operated exclusively within a cloud-based sandbox — an isolated online environment with its own file system, browser, and command line. With “My Computer,” the agent steps beyond those boundaries and gains direct access to local files, applications, and resources on the user’s machine. Control is exercised via terminal commands that Manus executes autonomously.
Possible use cases range from simple file organization tasks to full-scale software development. The company cites examples such as automatically sorting thousands of photos into categorized subfolders or bulk-renaming invoice files according to a uniform naming scheme. Tasks that would take hours manually are meant to be completed in minutes.
In addition, Manus can access all locally installed development tools via the command line, including Python, Node.js, Swift, and Xcode. The company demonstrated this by developing a complete Mac app for real-time meeting translation, built in around twenty minutes without manually opening Xcode or writing a single line of code by hand.
Leveraging local computing power
Another focus of the new feature is making use of idle hardware. Users can instruct Manus to deploy the local GPU for training machine learning models or for inference tasks. An always-on machine — such as a Mac mini — can thus be turned into a round-the-clock AI assistant accessible from any device anywhere in the world.
Manus combines cloud intelligence with local computing power. This makes it possible to assign a task from a smartphone that the home computer then carries out autonomously while the user is on the go.
Integration with existing services
“My Computer” complements Manus’s existing integration with services such as Google Calendar and Gmail. A concrete scenario: a user urgently needs a contract file from their home computer while away. Manus can remotely locate the document and send it directly to the recipient via the linked Gmail account. Local file and cloud service are seamlessly connected in the process.
In addition, recurring tasks can be set up as scheduled routines — for example, a daily cleanup of the downloads folder or the weekly generation of a summary report from local data.
Security and control
Given the far-reaching access permissions, Manus explicitly emphasizes the aspect of user control. Every terminal command requires the user’s explicit approval before it is executed. Two modes are available:
- “Always Allow”: For trusted, recurring tasks that require no manual confirmation
- “Allow Once”: For a single, controlled execution with prior review
“You are the commander; Manus is the executor. That relationship will never change.” (Manus, official announcement)
Parallels to OpenClaw
Numerous observers see clear similarities between “My Computer” and OpenClaw, the personal AI agent created by Peter Steinberger. Steinberger recently transferred OpenClaw to a foundation and joined OpenAI, where he is tasked with advancing the topic of AI agents for end users. Meta CEO Mark Zuckerberg had also attempted to recruit Steinberger for his company, but lost out to OpenAI CEO Sam Altman.
Both products promise personal AI agents with persistent memory that can independently plan and execute complex tasks. Manus positions itself as the lower-barrier alternative: while OpenClaw reportedly involves more complex setup processes, security risks, and unpredictable costs, Manus focuses on ease of use through familiar interfaces such as Telegram, and in the future also WhatsApp, Slack, and Discord.
Meta’s strategy behind the acquisition
The purchase of Manus for more than two billion dollars, completed at the end of 2025, is part of a broader realignment of Meta in the AI space. The company is under considerable pressure: its own Llama model series is considered non-competitive against the offerings of OpenAI, Google, and Anthropic; there is even talk of licensing Gemini in the interim.
Adding to this is the departure of AI chief Yann LeCun and the multi-billion-dollar acquisition of Scale AI, which has so far produced no visible turnaround.
With Manus, Meta is not buying its own language model, but a successful business model. Manus is built on third-party models — including offerings from Anthropic, Alibaba (Qwen), and OpenAI — and sells the agent-based orchestration layer as a subscription service to small and medium-sized businesses. Just eight months after its launch in March 2025, the startup reached an annualized revenue of 100 million dollars, with consistent monthly growth of over 20 percent.
In the long term, Meta plans to integrate Manus technology into its own products and scale it across the billions of users on its platforms; connections to Meta’s advertising services and Instagram have already been established.
The goal is clear: the company no longer wants to be left behind in the race for autonomous AI systems capable of taking over complete workflows. Whether the bet will pay off remains to be seen, as Manus has no proprietary model capabilities and therefore remains dependent on external providers.
