Meta Platforms has launched Muse, an AI agent designed for individual users, marking the latest step in Mark Zuckerberg's push to integrate artificial intelligence into daily life and a major bet in the consumer AI assistant race.

Muse is now available to users in the United States as of Tuesday, capable of handling everyday tasks such as online shopping, purchasing movie tickets, booking golf lessons, and even completing school activity permission forms. According to Meta's Chief AI Officer Alexandr Wang, the product's core design principle is to be "extremely easy for everyday users to pick up."

In a lengthy policy essay published last month, Zuckerberg made his stance clear, describing consumer-facing personal AI assistants as Meta's largest commercial opportunity. He outlined his vision in the document: "Everyone will have a highly capable personal agent that knows you, your goals, and everything you care about."

Meta's ultimate objective is to bring Muse to its billions of existing users across Instagram, Facebook, and WhatsApp. The company's shares closed down 0.53% on Tuesday.


Core Capabilities: From Handling Tasks to Managing Goals

Muse is designed to function as an agent tool that requires no technical background to operate. Users can issue instructions through a dedicated Muse app or directly within WhatsApp via conversational prompts, and the assistant can autonomously execute actions, including opening browsers, filling out forms, sending emails, and even negotiating. The product is powered by Meta's latest flagship model, Muse Spark, which has been purpose-built for real-world autonomous task scenarios.

For longer-running tasks, Muse can continue functioning even after a user closes the app, proactively checking back when progress is made or when user authorization is needed, such as before sending an email or completing a purchase. The assistant also demonstrates cross-app memory and contextualization. For instance, a recipe video a user saves on Instagram can be automatically converted into a shopping list, and if a user mentions a friend's dietary restrictions, Muse will factor that into invitations before sending dinner plans. Meta positions this feature as a proactive perceptiveness that "remembers key points with a single mention."


Architecture and Privacy: Cloud-Based Operation with User-Exclusive Data Access

In terms of privacy and security architecture, Meta has built a dedicated virtual machine system for Muse called Muse Secure VM, which strictly isolates user data from other users. The company also announced it will roll out Muse Confidential VM later this year, at which point user data will be encrypted with keys held only by the user, making it inaccessible even to Meta itself. Wang described this as an "extremely secure architecture," noting that while Muse retains operation logs and pending plans, it does not share data with Meta's advertising systems and proactively confirms with users before handling sensitive actions. For non-sensitive matters, Muse can operate independently

. Wang has named his own personal agent Euler, after his dog, using it to identify and purchase furniture for a new home, plan meals, and design workout routines. "I genuinely use it as a second brain," he said. Users can also customize the agent with their own names and avatars, set its communication style, and connect it to email, fitness trackers, and other apps so it can continuously push forward on user-defined goals in the background.


Ecosystem Integration: From Ticketmaster to the Google Suite

Muse has already established connections with several mainstream service platforms, including Ticketmaster, OpenTable, and Google's Gmail, Calendar, and Drive. It is also deeply integrated into Meta's own ecosystem, seamlessly working with Instagram and WhatsApp. David Singleton, who leads personal agent engineering, joined Meta earlier this year after previously serving as Stripe's CTO and CEO of AI startup Dreamer. He stated that if users use a service that has an API, "you can directly have Muse build a custom connector for you," while the team continues to expand official partnerships and service integrations. Muse is powered by Meta's latest AI model, Muse Spark 1.3, and will later be integrated into Meta's AI smart glasses.


Strategic Position: The First Step Toward 'Personal Superintelligence'

Meta described personal superintelligence in its launch statement as "one of the most transformative technologies of a lifetime" and framed Muse as the first step toward that vision. In terms of market strategy, Meta has chosen to embed Muse directly into WhatsApp, a messaging platform with a massive user base, aiming to drive adoption without requiring users to alter their habits. On the business model front, Muse offers both a free tier and paid plans, with monthly fees of $20 and $100 for access to more computational power. Wang and Singleton anticipate most users will stay on the free version, with paid tiers targeting users who have more advanced needs.

On the talent front, Meta's Superintelligence Labs division continues to aggressively recruit top AI engineers. Last year, Meta poached Ruoming Pang, a senior executive from Apple's AI model team, and brought on several notable researchers from OpenAI. Singleton also joined this year to oversee Muse's engineering development. Meta also previewed that Muse will launch its Confidential VM version later this year, featuring end-to-end encryption for enhanced privacy protection, which could serve as a key differentiator in the AI agent security landscape.