Meta Expands Personal AI Muse to Japan

Meta began offering its AI agent Muse in Japan on the 23rd and expanded it to Singapore on the 27th, local time. Vishal Shah, vice president of Meta’s AI division, described a system designed to understand users’ goals and preferences and help with daily life without waiting for instructions. Meta says it has pursued “personalized superintelligence” in developing Muse for about a year and a half.

Shah said Muse checks his calendar, his children’s schedules and emails from their school at 6 a.m. each day, then creates and prints a family newspaper combining local news, sports, riddles and jokes. He said Muse gathers information from multiple services and organizes it into material relevant to the family, without requiring detailed instructions every morning. Users can give Muse a name and personality and use it as a personal agent.

Meta said advertising that helps people discover products and brands serves a different purpose from Muse, which helps users when they are considering a purchase. As an example, the company said Muse could help find a backpack a user discovered several months earlier, drawing on that past interest. For monetization, Meta is considering shopping and e-commerce integrations and subscriptions, while first making Muse broadly available for free and working to control the costs of inference and computing resources.

Muse combines connectors for external services with virtual machines, a browser and code execution. Meta says it can create and run code on a virtual machine or access services through a browser or API even when a dedicated connector is unavailable. At launch, Muse can use information from Instagram, Facebook and Messenger. Meta also sees uses such as finding a cafe discovered while traveling or a recipe to try later, and plans to expand shopping and delivery integrations according to regional needs.

Meta released the Muse Connector Platform on September 18, allowing external developers to submit connectors for use with Muse after Meta reviews them for functionality and security. The company cited connections to shopping and banking services as possible uses. In October, it also released Muse Gadgets, an open-source project for linking Muse to custom devices built with Raspberry Pi or ESP32.

Shah said conversations with Muse are not passed to Meta’s advertising business, and interactions with virtual machines are not shared with advertisers. Users on the free plan can opt out in settings of having their data used to improve AI models, and the app has a lock feature such as biometric authentication. Meta said it chose Japan and Singapore as expansion markets after the United States because many users there adopt and use new technologies and because their regulatory environments support innovation. The company also plans to focus on models tailored to local needs, such as finding restaurants or making suggestions around school events, and to expand connections to external services and devices.

Meta began offering its AI agent Muse in Japan on the 23rd and expanded it to Singapore on the 27th, local time. Vishal Shah, vice president of Meta’s AI division, described a system designed to understand users’ goals and preferences and help with daily life without waiting for instructions. Meta says it has pursued “personalized superintelligence” in developing Muse for about a year and a half.

Shah said Muse checks his calendar, his children’s schedules and emails from their school at 6 a.m. each day, then creates and prints a family newspaper combining local news, sports, riddles and jokes. He said Muse gathers information from multiple services and organizes it into material relevant to the family, without requiring detailed instructions every morning. Users can give Muse a name and personality and use it as a personal agent.

Meta said advertising that helps people discover products and brands serves a different purpose from Muse, which helps users when they are considering a purchase. As an example, the company said Muse could help find a backpack a user discovered several months earlier, drawing on that past interest. For monetization, Meta is considering shopping and e-commerce integrations and subscriptions, while first making Muse broadly available for free and working to control the costs of inference and computing resources.

Muse combines connectors for external services with virtual machines, a browser and code execution. Meta says it can create and run code on a virtual machine or access services through a browser or API even when a dedicated connector is unavailable. At launch, Muse can use information from Instagram, Facebook and Messenger. Meta also sees uses such as finding a cafe discovered while traveling or a recipe to try later, and plans to expand shopping and delivery integrations according to regional needs.

Meta released the Muse Connector Platform on September 18, allowing external developers to submit connectors for use with Muse after Meta reviews them for functionality and security. The company cited connections to shopping and banking services as possible uses. In October, it also released Muse Gadgets, an open-source project for linking Muse to custom devices built with Raspberry Pi or ESP32.

Shah said conversations with Muse are not passed to Meta’s advertising business, and interactions with virtual machines are not shared with advertisers. Users on the free plan can opt out in settings of having their data used to improve AI models, and the app has a lock feature such as biometric authentication. Meta said it chose Japan and Singapore as expansion markets after the United States because many users there adopt and use new technologies and because their regulatory environments support innovation. The company also plans to focus on models tailored to local needs, such as finding restaurants or making suggestions around school events, and to expand connections to external services and devices.