Meta Launches Muse AI Agent to Automate Tasks Across Apps and Websites 24/7
Updated
Updated · KDnuggets · Oct 5
Meta Launches Muse AI Agent to Automate Tasks Across Apps and Websites 24/7
3 articles · Updated · KDnuggets · Oct 5
Summary
Meta on Sept. 8 launched Muse in the U.S. as a personal AI agent that can browse websites, connect apps, send emails, make purchases, fill forms and keep working after users close the app.
Muse runs on the Muse Spark 1.3 reasoning model and an isolated Linux virtual machine for each user, letting it manage long-running goals, use tools, spawn subagents and maintain persistent state.
Sentinel acts as a separate permission layer, deciding whether actions are allowed, blocked or sent for user approval as Meta tries to limit prompt-injection, payment and privacy risks.
Meta made Muse available on iOS, Android, the web and WhatsApp, and said early usage beat internal projections; the service quickly climbed U.S. App Store rankings.
The launch extends Meta's broader Muse push after it also unveiled the Muse Charm device, while highlighting a wider shift from question-answering chatbots toward consumer AI agents that take actions.