Meta Muse: Tasks, Pricing, Privacy, and Trust Explained

Muse is Meta’s personal AI agent: software intended to complete work for a person, rather than simply answer a question. Meta introduced it for U.S. users on September 8, 2026, positioning it as an agent that can work across connected apps and services and carry out tasks after a user delegates them. 1719

What can Meta Muse do?

Muse is designed for everyday digital errands and longer-running projects. Meta and reporting on the launch say it can browse websites, fill out forms, send emails, create plans, turn recipe Reels into grocery lists, manage bills, make purchases, and book travel. It can also help with restaurant and event-related tasks through connected services. 171920

The important distinction is execution. A user can give Muse a goal, then allow it to take the required steps inside authorized accounts. It can continue delegated work when the app is closed, while pausing to request approval for actions considered sensitive. 320

Where Muse is available and what it costs

At launch, Muse is available in the United States through iOS, Android, and the web. Meta has also described access through WhatsApp, with support for its AI glasses planned for later. 2317

Pricing is structured as a metered free tier plus two subscriptions:

• Free: usage-limited access
• Power: $20 per month
• Maximum: $100 per month 23

Availability and usage limits may change during the rollout, so the live product page is the best place to confirm eligibility and plan limits.

How Muse connects to email, calendars, payments, and other services

Muse is meant to work through services that a user chooses to connect. Reported categories include email, calendars, payment services, shopping, dining, travel, health and fitness, smart-home, music, and event apps. 19

That access is what makes the product more capable than a standalone chatbot—and more consequential. An agent that can draft an itinerary but cannot access a calendar or booking service is limited; an agent that can access both may be able to complete the task. Users should therefore treat each connection as a separate decision and grant only the permissions necessary for the tasks they actually want automated.

Payments: Link by Stripe, with Shop Pay and 1Password planned

For purchases, Muse can use Link by Stripe. Meta says this flow can use a one-time-use card so the agent can complete checkout without exposing a user’s actual card details to the merchant. 2022

Meta has also said that Shop Pay and 1Password support is planned. Shop Pay would expand checkout options, while 1Password could enable the agent to use existing logins through a password-management integration. 1822

This does not eliminate transaction risk. It changes the way credentials and payment details are handled, while a user still must decide whether to authorize an agent to buy something or act within an account.

Meta’s security and privacy design claims

Meta says Muse runs on Muse Secure VM, a dedicated cloud computer with its own browser. The company describes Sentinel as a separate supervisory agent that gates external network and connector activity. In practice, Meta says Muse must obtain Sentinel authorization before it can reach the internet or use connected tools. 172021

Meta’s stated safeguards include:

• Per-user isolation: each person’s work runs in a dedicated cloud virtual machine. 217
• Approval gates: sensitive or high-risk actions are intended to pause for a user decision. 320
• Credential separation: Meta says the agent does not directly see raw passwords or payment methods; reports describe surrogate credentials or secure storage being used instead. 121
• Advertising separation: Meta says Muse conversations and VM data are not shared with its advertising systems. 425
• AI-training choice: Meta says users can opt out of having their data used to train its AI models. 2
• Confidential VM roadmap: Meta is working on a confidential version intended to prevent Meta from seeing activity within the virtual machine. 21

These are significant design claims, but they should be read precisely: they describe Meta’s intended controls and architecture. They are not, by themselves, independent audit results or a guarantee that an agent will never make an error, encounter a malicious webpage, or be misused through an overly broad permission.

Why trust is the real product test

Muse asks for a level of access that ordinary chatbots generally do not need: email, schedules, shopping behavior, reservations, travel plans, payment flows, and potentially the authority to spend money. That makes reliability, revocable permissions, clear approval prompts, and data minimization core user requirements—not secondary features.

The launch also arrives amid scrutiny of Meta’s broader consumer-protection and privacy history. CNBC reported that Meta reached an $18 billion settlement in a multistate case concerning alleged social-media harms, and coverage of Muse has explicitly raised the question of whether consumers will trust Meta with this new kind of access. 3319

Past controversies do not demonstrate that Muse is unsafe. They do, however, make it reasonable for users to demand a higher standard of transparency, independent security testing, understandable permission controls, and a practical way to disconnect services or delete data.

How to approach Muse safely

For anyone trying an agent like Muse, the sensible starting point is low-risk work: gathering options, creating lists, or preparing a draft email. Add account connections gradually, review every approval request, and avoid granting payment or inbox access until its value is clear.

The appeal of Muse is straightforward: fewer repetitive online chores. Its adoption will depend on a harder question: whether the promised safeguards hold up when an agent is acting across the accounts, purchases, and personal information people rely on every day. 1719