Meta's Muse AI agent launches in Japan and Singapore

Meta's pitch for Muse rests on reversing the usual chatbot relationship: rather than responding to prompts, the agent is meant to infer a user's goals and act on them, a concept Meta says it has been developing for about a year and a half under the banner of "personalized superintelligence." Shah's family-newspaper example illustrates the intended texture of that shift — pulling from calendars, school emails, local news and sports, then producing something a household actually uses, all without daily instructions.

The commercial logic Meta describes is layered. Muse is meant to sit in the research-and-consideration stage of shopping, while Meta's advertising business keeps the discovery stage, so the two are framed as complementary rather than cannibalistic. Monetization is left open: free access first, with e-commerce-related opportunities and a possible move to subscriptions under consideration, alongside efforts to reduce inference and compute costs.

Underneath the agent, Meta is assembling more than a model. Muse combines virtual machines, a browser and code execution, and can reach external services through APIs rather than only purpose-built connectors — with shopping and banking cited as examples. Meta's own apps such as Instagram, Facebook and Messenger feed it from launch, and Meta says it will add connectors with regional needs in mind, since actual usage varies widely by country. The developer platform opened on September 18, and an October open-source project, Muse Gadgets, extends Muse to homemade devices built on Raspberry Pi and ESP32. Japan and Singapore were chosen, according to Meta, for users eager to adopt new technology and for regulatory environments that support innovation.