TL;DR: Meta launched Muse on September 8, 2026, as a U.S.-only personal AI agent for adults that acts on email, shopping, travel and long-term goals.

Meta launched on Tuesday a personal artificial intelligence agent called Muse for people 18 and over who want help with day-to-day tasks like schedules, shopping and turning long-term goals into action plans, according to The Associated Press. The parent company of Instagram and Facebook said the product is only available in the U.S. for now and stressed safety and privacy features built around a dedicated, secure virtual machine that houses both the agent and user data. People can message Muse in a separate app called Muse or message it in WhatsApp, where it can handle simple jobs like sending an email or booking travel and larger projects like creating a yearlong exercise plan or setting up a new business.

What Muse is supposed to do

Meta described Muse as a step beyond a chatbot that only answers questions, saying it develops a personalized plan once a person shares a goal and then advances the work on its own, according to The Associated Press. The company said in a blog post that the agent can open a browser, fill out forms and negotiate on a user’s behalf. That description matches the industry definition of an AI agent cited by The Associated Press: a program combined with a large language model that can search for knowledge and complete tasks without explicit, step-by-step instructions, such as receiving a message from a coworker, figuring out a likely reply and sending the response.

Meta said the Muse app lets people offload digital tasks like booking appointments, filling out electronic forms and even monitoring home security camera feeds to AI-powered assistants, according to CNBC. Meta AI chief Alexandr Wang told CNBC the company designed the app so it feels approachable and friendly and explainable without feeling too complicated, while behind the scenes it might run advanced coding workflows, build integrations and do heavy lifting while keeping the experience simple. The agent is designed to be more proactive and long-running than typical chatbots, allowing users to name their agent, create an avatar and customize how it communicates, according to Axios.

How to reach it and what it costs

Muse is rolling out for iOS and Android users in a dedicated Muse app and through the website Muse.ai, and users can also message Muse directly in WhatsApp, according to WIRED. Initially Muse is available in the U.S. and works on iOS, Android and the web, with support coming soon for Meta’s AI glasses as a future addition, according to Axios. Users of Meta’s AI glasses will soon be able to interact with the Muse agent as well, a connection Meta described as planned rather than live, according to WIRED.

Meta is offering a free tier of Muse, as well as two subscription options at $20 per month and $100 per month, according to Axios. Wang said the Muse personal agent will be available in either a free tier or through monthly subscription plans of $20 or $100 depending on usage, according to CNBC. Wang told Axios that for the vast majority of users, they should be able to do what they need within the free tier, while the paid tiers help cover compute costs for power users. People can try Muse for free, but users who want to automate large volumes of digital tasks will need one of the company’s AI subscription plans, according to WIRED.

Users of the Muse personal agent app will be able to see a feed that shows updates from connected Facebook and Instagram accounts or articles and stories from the web, depending on what they want the feed to aggregate and summarize, according to CNBC. That feed and an ideas tool that recommends agent tips and tricks are not in the WhatsApp version, although the core conversation capability is the same across the services, according to CNBC. There is no advertising within Muse, but Wang said the company is exploring commerce opportunities that could generate additional revenue, according to Axios.

Secure VM and Sentinel controls

Each Muse user shares a dedicated computer in the cloud with their agent, where the agent lives and where data and credentials for connected services are stored, according to the Meta AI Research blog. Each virtual machine is an isolated Linux box with a browser and enough storage, CPU and memory to do real work like compiling code the agent writes, developing custom skills and handling concurrent sub-agents and scheduled tasks. Muse clients on iOS, Android and the web connect directly to that virtual machine through a secure transport layer.

The virtual machine is structured to separate what the user and agent do from built-in protections, using Linux isolation primitives, according to the Meta AI Research blog. The internal name for Muse in the codebase is Hatch, and the Hatch daemon plus the workspace filesystem and tools operate in a runtime container where root inside the cell maps to an unprivileged host user, with its own filesystem, virtual network interface, filtered system calls and limited kernel capabilities. Security-sensitive services run outside that runtime cell because the cell is expected to process untrusted data, including a safety service that inspects requests and responses, privilege-separated workers that run connector code, and an authentication service for credential storage.

A separate system called Sentinel controls Muse’s access to the internet and connected services, according to Axios. Muse can suggest an action, but Sentinel determines whether it is allowed, blocked or sent to the user for approval. People can distinguish between read and write access, including more granular permissions within a service, and permissions can be narrowly scoped to a particular task, service, transaction or period of time. Meta executives said they were mindful that asking people to approve every small action can lead to reflexive approvals that undermine safety, so Muse is designed to require approval for more sensitive actions while allowing previously authorized, lower-risk tasks to proceed.

Payments, connectors and human checks

Muse is most useful when connected to email and other systems like a car or home, with an initial set of connectors to third-party systems and to Meta apps like Instagram and Facebook, according to the Meta AI Research blog. For each connector, Meta said it worked with the service provider to integrate the application programming interface and wrote detailed SKILL instructions on how to get the most from each connector. Muse can also write its own custom connectors for other services when they have their own APIs or command-line interfaces. Users can connect Muse to apps including email, calendar, payments, health and smart home, according to Reuters.

Muse is also capable of making purchases on behalf of users, checking out using payment infrastructure designed by Stripe, according to WIRED. The payment tool, which Stripe calls Link, issues a single-use card number so agents do not enter a person’s real financial information across the internet. Meta said Muse is the first AI agent covered by Link’s purchase protections for agents, which guarantee no-fee returns. Wang said the app runs within its own isolated environment inside Meta’s computing infrastructure and never sees actual passwords or payment details and asks before doing anything sensitive, according to CNBC.

David Singleton, Meta Superintelligence Labs vice president of engineering for consumer products, said Sentinel looks at everything moving out of the virtual machine and either matches it to an existing permission or presents a human-in-the-loop dialog to ask for approval, according to WIRED. Singleton said those check-in prompts come directly to the user and are not filtered through the model, a design intended to guard against prompt-injection attacks. The Meta AI Research blog said every interaction with the outside world runs through a Sentinel that the agent cannot override, and that the harness does not see real credentials.

Training, privacy options and bug bounty

By default, queries made to one of Meta’s models are able to be used by the company, but there is an option to turn that off, according to Axios. Users must opt out of letting Meta use their interactions with the Muse personal agent to train the company’s AI models, and if they do not opt out the company will scrub critical personally identifying information before using conversations to improve models, according to CNBC, citing Singleton. While Secure VM is built to maintain security and privacy, it is not a fully locked box because Meta is barred by policy from accessing user Muse data but access would still be technically possible, according to WIRED.

Meta said it is working on a confidential version where Meta cannot see what takes place inside the user’s virtual workspace, with release aimed before the end of the year, a future plan rather than a current feature, according to Axios. The company trained its model with a focus on zero-shot tool calling using command-line interfaces and skills, long context, long-trajectory instruction following with awareness of prompt injection, and multi-agent coordination, according to the Meta AI Research blog. Meta said it hardened Muse based on internal use, agentic red teaming and issues found by security researchers in a private bug bounty program, and it opened the Muse bug bounty to anyone with awards up to $300,000 for valid reports, including up to $130,000 for successful prompt-injection attempts affecting one user.

Why Meta says Muse matters now

For Meta, the agent fits into CEO Mark Zuckerberg’s vision of AI superintelligence available to everyone, which he outlined in a 6,500-word essay published online the prior month, according to The Associated Press. In that document, Zuckerberg said the company is working toward an era where everyone will have tools to create new businesses, receive Ph.D.-level tutoring and receive personalized lifestyle tips. Zuckerberg wrote that everyone will have an exceptionally capable personal agent that understands the user and goals and works around the clock on relationships, health, career, finances, home management and hobbies.

The app, given the internal code name Hatch, is powered by the Muse Spark family of foundation models that Wang has been introducing since April, according to CNBC. Zuckerberg hired Wang last year as part of a $14 billion investment in his startup, Scale AI, and has told investors that supercharged digital assistants that sift through emails or look for furniture deals represent the next big leap for AI models, according to CNBC. In the company’s earnings call in July, Zuckerberg said new personal agents will be the foundation for the next wave of products and revenue lines in the months and years ahead.

Meta said Muse was designed so there is no learning curve and users can prompt it in natural language to send emails, book travel or help sell a car, according to WIRED. Muse is Meta’s latest attempt to compete with AI agents such as OpenClaw and Instinct that users message to automate digital tasks, as a product of Meta Superintelligence Labs, the AI unit Zuckerberg formed roughly a year ago to catch up with OpenAI and Anthropic, according to WIRED. Wang said the Muse personal agent is intended to be more accessible to a broader audience than competing products, while conceding it is still pretty early in this new era of personal agents, according to CNBC.

Sources referenced