Meta has introduced Muse Spaces, custom interactive experiences created by describing an idea to the Muse app, such as a gift tracker, school dashboard or trip planner. The 6 November product announcement says users can keep a space in the sidebar or main tabs and connect apps so its contents update in the background.
The change makes a recurring task into something a user can return to inside Muse, rather than starting another conversation each time. Meta also announced specialised model upgrades alongside Spaces. Those are related developments in the same product, but a new interface and a model’s evaluation results describe different things: the research figures are not a promise that every Spaces user receives identical improvements.
Muse Spaces can move between the sidebar and main tabs
A description is the starting point, not the final specification. Meta says users can keep adjusting a space through conversation, changing its appearance, adding missing elements or altering how it works. They can hide spaces that are not currently needed and schedule when they appear, for example moving a weekly planner out of the way at the weekend.
The company’s examples range from a music tracker and trivia game to a client hub containing projects, deadlines and invoices. A school dashboard could find relevant events in school emails and add them to a calendar; a trip planner could organise flights, hotels, reservations and a daily itinerary. These are Meta’s proposed uses, not a claim that every connected service supports every example.
Connecting apps is intended to keep information current without another prompt. Meta says users can also check items off, add notes and edit directly inside a space, with those changes available when they return. The company says Spaces is available everywhere Muse is available, while sharing spaces with other people remains something it is exploring.
For the geographic rollout context, TSN’s Muse launch coverage for Japan and Singapore follows the agent’s expansion beyond North America. The Spaces announcement adds a feature within existing availability rather than announcing another named country launch.
Specialised models bring memory and recovery claims of their own
In a separate research post, Meta says specialised versions of its existing models were trained using Muse Wonder and are rolling out to Muse and Meta AI. This is distinct from the earlier Muse Spark Open weights release; Spaces is an in-app experience, not a new downloadable model-weight package.
Meta reports a Personal Superintelligence Index score of 47% for the current Muse model versus 38% for the previous one. Its Active Memory evaluation reports 90% versus 76% on tasks correctly using critical context. The company describes Active Memory as recognising when past information matters, distinguishing temporary from lasting details and identifying information that has become stale.
The post also reports a 2.4-fold reduction in forgetting critical information and a 67% improvement in completing everyday tasks when unexpected issues arise. It connects the latter behaviour to Active Focus, which is intended to help a model recover or find alternatives when a task runs into a bottleneck. These are company-reported evaluations; the post does not supply absolute task-completion rates for the 67% claim, which is not a 67% success rate or a 67-percentage-point gain.
A persistent workspace tests more than a benchmark score
A space’s practical value depends on the relationship between its layout, the information feeding it and the task someone repeatedly needs to perform. Better memory could help preserve relevant context, but it does not by itself establish that the right update appears in the right panel or that a particular connected-app workflow completes reliably. The product announcement explains how people organise their experiences; the research post describes selected capabilities of the models behind the wider services.
That separation matters when comparing progress. A repeatable evaluation can make a memory improvement visible, while a personal workspace brings together changing information, user edits and different routines. The same score cannot specify the result for every combination of those elements. Nor does the presence of specialised models establish a particular deployment schedule for every user or every task inside a space. The clearest reading is a product feature arriving alongside a model rollout, not one universal performance result attached to each custom experience.
Connected-app use also makes trust in the underlying agent relevant. TSN’s Muse bug-bounty coverage describes the company’s separate account of finding and fixing security weaknesses. That security work, interface customisation and task benchmarks answer different questions about the product.
For now, the next product boundary is sharing: Meta is exploring collaborative spaces, but the launch described here is a personal experience inside the existing Muse app.
Sources
- Meta Newsroom: Design the Muse App, Just for You, 6 November 2026
- Meta AI Research: Bringing Specialized Models to Muse and Meta AI, 6 November 2026
Related stories
Meta Launches Muse AI Agent in Japan, First Stop Outside North America tracks availability and regional connectors, useful context for a feature Meta says is arriving wherever Muse is available.
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
