Meta Muse Spaces: Build Your Own App Inside Muse, Plus the Unverified 67% Claim

Meta introduced spaces in its Muse app on November 6, 2026: you describe an idea, and Muse builds a small custom app for it. Meta's newsroom post says spaces are live "everywhere Muse is available." The same day, Meta said it is rolling out specialized models trained with learnings from Muse Wonder, and reported two large improvements. Those numbers have no published method yet, so this article separates what Meta confirmed from what is only a claim.

TL;DR

What is a space?

Meta defines a space as "a custom experience you create inside Muse by describing your idea." Its examples are a holiday gift tracker, a daily word game or a planning board for a DIY project. Your first prompt is a starting point. You then adjust the space by chatting with Muse: add what is missing, try a new look or change how it works.

Placement is part of the feature. You can add spaces to the sidebar, move them into the main tabs, hide ones you do not need and schedule when they appear. Meta's example is a weekly planner up front during the workweek, swapped for weekend plans on Saturday.

Video: Meta

What does Meta say you can build?

The newsroom post lists nine ideas. Each ties to data Meta products hold or that you connect.

Meta adds that if you connect the apps you use, your spaces "keep themselves up to date in the background," so new concert dates, price changes or savings progress appear without asking. You can check things off, add notes and edit inside the space.

Why does this matter?

Spaces turn a chat assistant into a place where results persist. Instead of asking Muse the same question each morning, you keep a view that updates. That is a step toward the personal agent Meta has been building. For the earlier pieces of that push, see our posts on the Muse personal agent launch and its VM security model, the Muse developer connectors platform, Muse business-in-a-box and Muse for Mac.

It also raises familiar questions. A space that reads school emails and Marketplace messages needs broad permissions. If you try one, connect only the apps it needs and review what Muse can see. Meta's post does not detail how spaces store data or how long, so we cannot say more about that.

How should you set up a space safely?

Every space that "keeps itself up to date" is a standing connection to your data. A short checklist helps before you build the first one.

• Start with one source. A trip planner needs email and calendar. It does not need your Marketplace messages. Connect the minimum.
• Name the purpose in the prompt. Tell Muse what the space may read and what it may not, then check the result. A model can ignore a soft instruction, so look at what the space actually shows.
• Review the connection list. After a week, open the connected apps and remove any you no longer use.
• Keep sensitive records out. Invoices, school emails and customer messages are personal or business data. Think about whether a consumer app is the right home for them.
• Delete what you drop. Meta says you can hide spaces. We found no statement in the post about deleting a space or the data behind it, so do not assume hiding removes anything.

Meta's own earlier Muse launch drew scrutiny over how its agent handled files and trust; our post on the Muse VM filesystem export covers one such question. If you run agents on your own machine and want a layer that blocks risky actions, AgentBeam, the agent security platform from the explainx.ai team, is built for that.

How does this compare with building an app by hand?

A space is closer to a personal dashboard than to software you ship. It lives inside Muse, uses Meta's connectors, and is shaped by chat. You cannot export it as code, as far as Meta's post says, and you cannot share it yet. That makes it useful for private routines and weak for anything you need to hand to a customer or a team.

For a business use such as an order tracker or client hub, the question is lock-in. Your workflow would live in an app you do not control, on a platform that can change features. For personal use the trade-off is easier: you get a working tracker in minutes and lose little if it goes away.

Developers who want more control have other routes in the same ecosystem. The Muse connectors and API posts above describe how to build agents with code. Spaces are the no-code end of the same stack.

What did Meta claim about the new models?

Meta's post says: "We're also rolling out specialized versions of our models to Muse, trained with learnings from Muse Wonder, to make your overall experience even better." It gives no names, regions or schedule.

A fuller account comes from a Digg summary of a post by the AIatMeta account the same day. Per that summary, Meta described two capabilities:

• Active Memory: recognizes when earlier information matters and separates what is still relevant from what has gone stale.
• Active Focus: recovers from setbacks or finds alternatives when constraints change.

For Muse, the summary reports Meta's figures as a "2.4x reduction in forgetting information critical to your goals" and a "67% improvement in getting unstuck as constraints evolve." For Meta AI, Meta described benefits without numbers: breaking complex problems into steps, asking sharper follow-up questions and adapting when users change direction.

The same summary says the post "did not identify the baseline model, evaluation methods, sample size or tasks used," and gave no region, account-tier or rollout detail.

Shears trimming a memory tree, picturing Muse Active Memory separating relevant from stale information

How do these relate to Muse Wonder?

Active Memory and Active Focus were the headline features of Muse Wonder, which Meta launched in October. See our Muse Wonder launch, pricing and benchmarks post for those. What is new on November 6 is that Meta says smaller, specialized models inherit the ideas and are coming to the consumer Muse app and Meta AI. TechCrunch covered that launch as Meta's first frontier model. Treat Meta's own figures with care until outside testers repeat them.

What is verified and what is not?

Note that an aggregator sometimes writes the 67% figure as "better at self-correcting and completing everyday tasks." The wording in the Digg summary is "getting unstuck as constraints evolve," which is narrower. We use the narrower form.

How can you test the claims yourself?

You do not need Meta's data to check the direction of the claim.

1. Memory test: give Muse a goal with three constraints in one session. Change one constraint a day later and see whether it applies the change and drops the stale one.
2. Setback test: ask for a plan, then say the first option is unavailable. Check whether it offers a real alternative or repeats itself.
3. Space test: build a simple space, such as a trip planner, and see how it handles an edit. Note how often it needs you to repeat instructions.
4. Repeat it: run the same test three times. One run proves nothing.

Report numbers, not impressions. If many people do this, a rough picture emerges before Meta publishes its method.

Bottom line

Spaces are a real, shipped feature that makes Muse more like a personal home screen. The accuracy claims for the new models are Meta's own, without baseline or method, and they are the part to wait on. Use spaces if the permissions are acceptable to you, and test the memory claims on your own tasks before you rely on them.

Details are from Meta's newsroom post of November 6, 2026 and a summary of Meta's social post, and may change.

Related reading

• Meta Muse Wonder Full: launch, pricing and benchmarks
• Meta Muse personal agent launch and VM security
• Meta Muse developer connectors platform
• Meta Muse business in a box
• Meta Muse for Mac
• Meta Muse Code out of beta