Summary: On November 6, 2026, Meta launched spaces in Muse — custom mini-experiences you create by describing an idea and then keep refining inside the app. Each space lives wherever you put it in the Muse interface, updates itself from the apps you connect, and can be edited just by asking. For builders, this is Meta turning a personal agent into a distribution surface where users assemble their own software, and it’s worth studying even if you never touch the product.
What a space actually is
A space is not a prompt or a saved chat. Per Meta’s announcement, you describe an idea — a holiday gift tracker, a daily word game, a planning board for a DIY project — and Muse builds it. The description is the starting point, not the deliverable: as you use the space, you keep adjusting it conversationally, adding what’s missing, changing the look, or reworking how it behaves.
Two mechanics make this more than a widget gallery. First, spaces persist. Every one you create becomes part of Muse, so you return to it whenever you like. Second, they can stay current on their own. You connect the apps you use, and the space updates in the background — new concert dates, price changes, savings progress — without you asking. You can still check things off, add notes, and edit directly inside it.
The example catalog Meta published is telling about the intended range. On the personal side: a shopping hub pulling gift ideas, Instagram-saved wishlists, and Marketplace listings into one view; a school dashboard that extracts events from your kids’ school emails and adds them to your calendar; a wedding planner with RSVPs, vendor invoices, and to-dos. On the work side: an order tracker for a small business showing new orders, customer messages, and what needs to ship today; a client hub for freelancers with projects, deadlines, and invoices. Plus purely fun ones like a trivia game about your favorite movies or sports team.
The interface bet: the app becomes a surface
The part I find most interesting isn’t the generation — it’s the placement control. You choose where each space lives in Muse: in the sidebar, in your main tabs, hidden when you don’t need it. You can also schedule when spaces appear. Meta’s example: keep a weekly planner up front during the workweek, then swap it for weekend plans on Saturday.
That’s a real product decision. It treats the Muse app less like a chat window and more like an operating system for personal software, where the user composes their own home screen from generated parts. If you’ve been watching agents move from conversations toward persistent workspaces, this is the consumer-facing version of that shift — related in spirit to what Meta’s agent models already do with memory and focus at the model layer.
How it connects to the rest of the Muse stack
Spaces sit on top of the platform Meta has been assembling since September. Muse itself launched on September 8, 2026 as a personal agent running in the US on iOS, Android, and muse.ai, free for most usage with subscription plans for heavier users. Architecturally, it runs on Muse Secure VM — a dedicated virtual machine holding both the agent and your data — with a separate Sentinel agent, isolated at the system level, that must approve anything before it reaches the internet. Muse checks with you before sensitive actions like sending email or making a purchase, and shows a complete audit trail.
Then on October 12, Meta shipped the Muse Wonder series, describing Muse Wonder and Muse Wonder Full as its most capable models, with Active Memory for carrying context across conversations and Active Focus for persisting through hard tasks. Wonder is available now through the Meta Model API and Muse Code, and Meta says it will soon power more of Muse, rolling out first to subscribers. The spaces announcement ties into this: Meta is also rolling out specialized versions of its models to Muse, trained with learnings from Muse Wonder.
So the sequence is deliberate: an agent with a secure runtime, stronger models behind it, and now a way for users to shape the surface. That’s a platform play, not a feature drop.
The trade-offs and what’s not established
A few things deserve skepticism. Sharing is not shipped. Meta says it is “exploring” ways for you to share spaces with others — so the collaborative angle exists only as a hint, with no timeline or mechanics stated.
The model upgrade path is also fuzzy. The Wonder rollout to Muse “first to subscribers” implies a tier split, but neither announcement quantifies subscription pricing or says whether spaces get Wonder-grade models at launch or only later. And while the Secure VM and Sentinel architecture covers Muse’s actions, the spaces post doesn’t specify how a space’s background refreshes from connected apps interact with those approval flows — which matters a lot if your space is pulling school emails or Marketplace messages automatically.
There’s also the usual caveat for anything described into existence: these spaces are generated software, and Meta hasn’t published anything about versioning, export, or what happens when a background edit breaks something you relied on.
What I’d do next
If you build agent products, spend thirty minutes in spaces this week with one concrete task — the small-business order tracker or freelance client hub examples are the closest to real workflows. Watch where the generated experience holds up and where it needs manual correction, because that boundary is your competitive gap.
If you’re building on Meta’s side of the fence, the Wonder models are already accessible through the Meta Model API and Muse Code, and the specialized-model rollout to Muse suggests the agent surface and the API are converging. The practical question to test: how much of what a space does can you replicate today via the API, and where does the built-in app surface win? Right now, the answer to the second half is mostly placement, scheduling, and connections to apps like Instagram and Marketplace — which is exactly the moat to watch.
Sources
- Design the Muse App, Just for You
- Introducing Muse Wonder and Muse Wonder Full: Frontier Models for Agents
- Introducing Muse: The World’s First Personal AI Agent Built for Everyone
View the original post on X