Meta's new AI agent Muse can shop, plan, and live in WhatsApp
Meta just launched Muse, a personal AI agent that can handle real-world tasks like booking travel, negotiating bills, and making purchases on your behalf. Powered by the new Muse Spark model, it works inside WhatsApp and dedicated apps while running on isolated cloud virtual machines to safeguard personal credentials. It is rolling out to US users on iOS, Android, and the web, with most core features completely free.

Meta wants to run your errands.
Meta just launched Muse, a personal AI agent designed to browse the web, book travel, send emails, and negotiate deals on your behalf. Powered by its new Muse Spark model, Muse is rolling out in the U.S. on iOS, Android, and the web, and is accessible directly inside WhatsApp.
A sandboxed computer with a digital babysitter
Delegating real-world chores requires handing over serious credentials. Meta’s answer is the Muse Secure VM, a dedicated cloud environment where each user gets an isolated virtual machine to store credentials and run browser tasks in the background after you close the app.
To keep the agent from going rogue, Meta placed a separate internal watchman on the machine called Sentinel. Muse cannot touch the wider internet without Sentinel’s green light, and sensitive actions like checking out or sending emails trigger direct user approval.
Payments are routed through Stripe’s Link wallet, which generates single-use virtual cards that keep your actual card number hidden while adding purchase protections for damaged goods or price drops. Integrations with 1Password and Shop Pay are slated to arrive next, along with an end-to-end encrypted "Confidential VM" later this year.
Chart climbs and competition
Early adoption shows quick traction, though not quite at a viral frenzy. According to Sensor Tower, Muse topped 83,000 iOS downloads in the U.S., climbing to No. 2 on the App Store charts. However, it trails the historic debuts of Threads (4.3 million launch-day installs) and ChatGPT, and sits at No. 338 in Google Play’s productivity category.
Meanwhile, rival agent startup Instinct, recently valued at $2.5 billion, is moving fast by launching agent-to-agent coordination so friends' bots can plan hangouts together.
There is also the elephant in the room: Meta dropped Muse just days after agreeing to an $18 billion legal settlement over social platform harms, leaving prospective users to weigh handy automation against the company’s lengthy record of privacy penalties.
Why it matters
For everyday consumers, Muse is one of the most frictionless entries yet into agentic AI. Instead of wrestling with developer setups or rigid enterprise dashboards, you can drop a task into WhatsApp and let the cloud machine grind through the web forms. For travel hunting or price tracking alone, it is worth testing for free right now.
For the broader industry, Muse shifts the competitive arena from raw reasoning benchmarks to packaging and distribution. If personal agents become our primary internet middlemen, Meta’s advantage lies in placing Muse inside the messaging feeds where billions of people already spend their days.
We think the Sentinel-monitored sandboxing and virtual-card checkout are smart engineering choices that realistically address safety limits while prompt injections remain unsolved. But the tension is obvious: Meta still notes it may access VM data for service support and uses interactions for model training unless opted out. Building a true everyday digital surrogate requires supreme trust, and convincing users to hand over their life logistics will be a steep climb for a social media giant.
Sources (3)
- about.fb.com(opens in a new tab)Originalabout.fb.com
- techcrunch.com(opens in a new tab)Reporttechcrunch.com
- theverge.com(opens in a new tab)Reporttheverge.com