Meta Unveils Muse: A Proactive AI Agent to Manage Your Digital Life
On September 9th, Meta officially launched its personal AI Agent, Muse. Positioned as a proactive assistant that helps users achieve goals and offers suggestions, Muse is powered by Meta's most advanced model, Muse Spark. It is accessible through a standalone Muse app, directly within WhatsApp, and is slated for future integration with AI glasses.
Unlike conventional AI assistants focused on conversational Q&A, Muse is fundamentally task-oriented. Users can describe what they need to accomplish in natural language, and Muse will formulate a personalized plan, coordinate time and resources, and execute specific tasks. Its capabilities include opening browsers, filling out forms, sending emails, booking travel, and even handling certain negotiations on the user's behalf.
For tasks that require more time, Muse can continue running in the background even after the user closes the application. It will send notifications when the task status changes or when user authorization is required. For instance, Muse will always request confirmation before performing sensitive actions like sending an email or making a purchase. Meta has highlighted use cases such as selling a car, negotiating to lower bills, and continuously adjusting a workout plan based on a user's changing schedule.
To handle payments securely, Muse has integrated with Stripe's Link service, becoming the first AI Agent to receive Link's shopping protection. Link generates single-use card numbers for Muse, preventing users' real card information from being exposed to websites. It also offers services like damage or loss protection, price drop compensation, and free returns for eligible items. Meta plans to integrate Shop Pay and 1Password in the future.
Muse also features long-term memory, allowing it to make proactive suggestions based on information provided by the user over time. For example, it can convert recipe videos saved on Instagram into a shopping list, recommend a menu for a party, and remember friends' dietary restrictions for future event planning.
Addressing the security concerns of an agent with long-term access to personal data, Meta introduced the Muse Secure VM. Each Muse Agent operates within an independent virtual machine in the cloud, where data and credentials for connected services are stored. A separate Sentinel Agent, isolated from Muse, audits its internet access and requests user authorization when necessary, ensuring a robust security framework.
Meta emphasizes that Muse itself cannot directly view user passwords or payment methods; these credentials are stored securely and called upon by the agent when needed. Users have full control over which apps Muse can connect to and can grant specific permissions like 'read' or 'execute,' which can be revoked at any time. Muse also provides a complete activity log of all tasks it has performed and plans to perform.
In terms of data privacy, users can opt out of having their interactions with Muse used to train Meta's AI models. Furthermore, conversations and data within the virtual machine will not be shared with Meta's advertising systems. Later this year, Meta plans to launch the Muse Confidential VM, which will encrypt the entire virtual machine—including data and conversations—with a key held exclusively by the user. Muse is currently rolling out in the United States for iOS, Android, and web users, with most basic features available for free alongside a subscription plan.