Three-file agent13 minAn inbox agent that labels every email and drafts replies in your voice, built in plain English inside Claude Code and then put on an hourly cloud schedule. Here's the four-level method, the three files that make it work, and the safety settings you shouldn't skip.Read now →

Meta's Free Muse Agent Now Builds You Custom Mini Apps While You Talk to It on the Phone
Meta just gave its Muse agent three upgrades in one week: Spaces that turn a sentence into a self-updating mini app, real-time voice calls, and memory and focus features that used to sit behind a paywall. Here's what they do, a hands-on style walkthrough, and the catches.
Picture this: you put your earbuds in, call your AI agent with your phone in your pocket, and start rattling off ideas on a walk.
A dashboard of every automation it runs for you. A calorie tracker that reads a photo of your lunch. A board game for a six-year-old who loves unicorns. By the time you're home, all of them exist as working mini apps.
That's now a realistic workflow in Meta's Muse app. In the space of two days in early November, Meta shipped three updates that change what its free agent is good for. And the most important one solves the problem that quietly made Muse frustrating for heavy users: chaos.
Here's what changed, how people are actually using it, and where it still falls short.
First, what Muse is
Muse is Meta's personal AI agent app, built by Meta Superintelligence Labs. It launched in the US in September 2026 on iOS, Android, the web and WhatsApp, and it can act across your connected apps and services rather than just chatting.
In October, Meta released Muse Wonder, its new frontier model, and said its headline features would come to subscribers first. That's the backdrop for why the November updates matter: several of those features have now been pushed down to everyone, for free.
The problem: one giant chat
Muse was designed around a single master thread, the main chat. That makes it simple to start. It also makes it messy once you hand it real work.
Users who leaned on Muse for daily tasks ran into the same wall:
- Everything piled into a few chats, and finding the right one got harder over time.
- Side chats helped, but it was easy to forget what each was working on.
- The list of scheduled tasks showed little detail. You couldn't easily see when a task last ran, when it would run next, or whether something had failed.
For a quick personal assistant, fine. For anything resembling a workflow, it broke down.
Update 1: Spaces, mini apps you describe into existence
On November 6, Meta introduced spaces. A space is a custom experience inside the Muse app that you create just by describing it.
Meta's own examples include:
- A shopping hub combining gift ideas, Instagram saves, Marketplace listings and buyer messages.
- A school dashboard that pulls kids' events out of school emails and adds them to your calendar.
- A music tracker for new releases and upcoming shows.
- A wedding planner with RSVPs, vendor invoices and to-dos.
- An order tracker for a small business, or a client hub for freelance work with projects, deadlines and invoices.
- A trip planner with flights, hotel, reservations and a day-by-day itinerary.
The key detail is that spaces aren't static. Connect your apps, and Meta says spaces keep themselves updated in the background, so new concert dates, price changes or savings progress show up without you asking. You can also check things off, add notes and edit inside a space.
Spaces are available everywhere Muse is available, and Meta says it's exploring ways to share spaces with other people.
What heavy users are building
The early-adopter use cases are more interesting than the official list. Some real examples of what spaces can do:
A command centre for your automations. This is the fix for Muse's biggest weakness. One user-built dashboard lists every scheduled task Muse runs, with a table showing the last run, how often each runs, whether it needs attention, and when it runs next. Suddenly you can see what your agent is doing.
Games. A simple brick-breaker clone. A tiny cooking game where your Muse avatar wanders a kitchen, picks up carrots, chops them and cooks them. These are toys, but they show the range.
An order tracker. A space that scans a Gmail inbox each morning for tracking numbers and order confirmations and collects them in one view. The work was already possible in chat; the space just makes the result glanceable.
A Black Friday price watcher. A list of wanted items that Muse monitors, flagging when one falls below a target price. In one example, a discount on a cell phone plan was already live while the other items still showed "no deal yet."
A trip hub. Every flight and every packing item for an upcoming trip in one place.
Update 2: Voice calls that actually feel like calls
The day before spaces, on November 5, Meta began rolling out real-time voice and expressive avatars in Muse.
How it works:
- Tap the phone icon in the top-right corner to start talking.
- It's rolling out on mobile, web and Mac, in the US first.
- Meta says you can interrupt mid-response, like a real conversation.
- There are hundreds of voices to choose from, from a New Yorker's energy to a pirate's flair.
- The avatar reacts while you talk and speak. You can go voice-only, keep the avatar alongside the chat, or go full screen.
Meta calls the underlying tech Muse Realtime Voice and Muse Realtime Avatar. Voice conversations are coming next to Meta's AI glasses and Muse Charm.
Lots of assistants have voice modes. What's different here is that Muse keeps working in the background while you talk. Meta's pitch is that you can describe what needs doing on a walk or with your hands full, and Muse handles it as the conversation continues.
A spaces-by-voice session, step by step
Here's what building several spaces in one phone call looks like in practice, based on one tester's session.
Request 1: an automation dashboard with personality. The ask: a table at the top showing each automation's last run, frequency, attention status and next run, plus a character for each automation, a variation of the user's Muse avatar wearing a different outfit and hat.
Then the user asked Muse for ideas to improve it. Muse suggested and folded in extras during the call:
- A speech bubble for each little character saying what it just did, like "checked at 9:13 and still nothing new."
- Characters that bounce or look sleepy depending on status.
- An "attention spotlight" for tasks that need the user.
Request 2: a photo calorie tracker. Snap a photo of food, have Muse analyse it, log the calories and track them over time. The user deliberately asked for a simple first version.
Request 3: a free-versus-paid report. As a space, with graphics and tables, comparing Muse's free and paid plans.
Request 4: a unicorn race game. For a six-year-old niece: a two-player board game where you roll dice and race unicorns to the middle. Muse's suggestions: one big roll button, a rainbow path to a castle in the centre, and letting the child pick her unicorn's colour.
Request 5: a morning report. Each day, check a content-schedule folder on the user's Mac mini for tasks assigned to them that day, and combine that with Google Calendar events.
Muse hangs up automatically after a stretch of silence, and you can also just tell it to hang up. The work keeps going after the call ends.
How the results turned out
- The comparison report arrived first, and it pulled in the user's own account data without being asked, such as how much of the weekly usage allowance had been consumed (23% in this case) and a banked bonus-token balance. It also noted that paid plans add more usage and access to Muse Wonder.
- The calorie tracker worked on a phone: a photo of a banana was correctly identified at about 100 calories and saved to a log.
- The unicorn game first came out bright, flat and functional. After a second voice request to make it look like a real 3D board game, Muse rebuilt it with a board you can spin around.
- The automation dashboard matched the earlier hand-built version closely.
- The morning report was the weak spot. Muse initially showed an empty report and struggled to find the right folder on the Mac. Once the folder was sorted out, it produced the overview, which on a quiet Sunday was just two items.
That last one is worth noting: Muse can read and write files on your computer and pull them into spaces, but only once you give it permission, and finding the right location isn't always smooth.
Update 3: Active Memory and Active Focus, now free
The third update is under the hood, but it may matter most.
When Meta launched Muse Wonder on October 12, two headline capabilities came with it:
- Active Memory: carrying important details across conversations, telling temporary information (a hotel for one trip) from lasting preferences (a dietary restriction), and noticing when earlier information has gone stale.
- Active Focus: staying on task when something goes wrong, trying alternatives or recovering from errors instead of handing the problem back to you.
On November 6, Meta said it's rolling out specialised versions of its models to Muse, trained with learnings from Muse Wonder. Reporting on Meta's announcement says these bring Active Memory and Active Focus to all Muse and Meta AI users at no cost.
Meta's own numbers, as reported by Data Studios:
| Measure (Meta internal) | Before | After |
|---|---|---|
| Personal Superintelligence Index | 38% | 47% |
| Active Memory eval, critical context used correctly | 76% | 90% |
| Forgetting critical information in Muse | Baseline | 2.4x less often |
| Everyday tasks with unexpected issues | Baseline | 67% better |
Two caveats. These are Meta's internal benchmarks, with no published methodology or independent check. And the free models aren't Muse Wonder itself. Wonder remains in paid tiers and on Meta's model API, priced at $4 per million input tokens and $20 per million output tokens, according to the same reporting.
Why it matters for spaces
Spaces that update themselves are only useful if the agent behind them remembers what you care about and doesn't give up when a website changes or a folder moves. Memory and persistence are exactly what make a background mini app trustworthy, which is why bundling these three updates together makes sense.
How to set up spaces the smart way
Create them
- In any chat, ask Muse to create a space and describe the app, game or dashboard you want.
- Or call Muse and describe several at once. Ask it for suggestions; it's surprisingly good at proposing small improvements.
Organise them
Your spaces live in the Library (on the left on desktop, at the bottom on mobile). The mistake most people will make is leaving everything there.
Instead:
- Find your most important spaces.
- Click the three-dot menu.
- Choose Move to sidebar to pin them for one-tap access.
Meta also lets you move spaces into your main tabs, hide ones you don't need, and schedule when they appear, such as a work planner on weekdays swapped for weekend plans on Saturday.
Automate them
Tell Muse how often a space should refresh when you create it, for example daily, several times a day, or twice a week. Muse may infer a sensible cadence on its own, but you can always go back to a chat and say something like: "Update my morning report every day at 6 a.m." Then you wake up to a handful of current spaces instead of a pile of chats.
Connect what you need
Spaces get more useful with data. Muse works with apps including Gmail, Google Calendar, Google Drive, Notion, OpenTable and Facebook Messenger, and one tester pointed out that connecting finances through Plaid could power a weekly or monthly spending-and-savings report as its own space.
Why this matters
The big AI companies are converging on the same idea: the chat window isn't the product. The product is an agent that does work in the background and shows you results in a useful shape.
Spaces are Meta's version of that, and it's aimed squarely at regular people rather than developers. You don't write code, set up a server or think about hosting. You describe a thing and it appears inside an app you already use, kept current by an agent that, as of this month, remembers more and gives up less.
And it's free. With billions of people already in Meta's ecosystem, that combination could put a self-updating personal dashboard in front of far more people than any developer tool ever will.
Counterpoints and risks
- Meta isn't seen as having the strongest models. Many power users still reach for models from OpenAI, Anthropic or other labs for high-stakes work. A sensible pattern is to use free Muse for personal automation and reach for a stronger model when quality really matters.
- Benchmarks are self-reported. The memory and focus gains come from Meta's own evaluations.
- Privacy trade-offs. Spaces that watch your inbox, calendar and finances mean handing an agent broad access. Meta says users can tell Muse to forget things, opt out of having interactions used for training, and disconnect services, and that Muse doesn't share conversations with its ad systems. Still, connect only what each space needs.
- US-first rollout. Voice is starting in the US, so availability will vary.
- Local files can be finicky. As the morning report showed, pointing Muse at the right folder may take some back and forth.
Practical takeaways
- Build the automation dashboard first. If you already run scheduled tasks in Muse, a status table showing last run, next run and anything needing attention is the single most useful space you can make.
- Use voice for brainstorming, text for precision. Voice is great for spinning up several spaces fast. Fine-tune in text.
- Start simple, then iterate. Ask for a basic first version, test it, then request upgrades, just like the 2D unicorn game that became a spinnable 3D board.
- Pin what you use daily. Move key spaces to the sidebar so they replace your chat list as your home screen.
- Set explicit refresh times. Don't rely on Muse guessing your schedule for anything time-sensitive.
FAQ
Is Muse free? Yes. Spaces are available to all Muse users, and the Wonder-trained models with Active Memory and Active Focus are rolling out to free users. Paid plans add more usage and access to Muse Wonder.
What's the difference between a space and a chat? A chat is a conversation. A space is a persistent mini app, like a dashboard, tracker or game, that lives in your Library or sidebar and can update itself in the background using your connected apps.
Can I use voice outside the US? Meta says voice and expressive avatars are rolling out in the US to start, on mobile, web and Mac.
Can Muse access files on my computer? It can read and write local files and use them in spaces, but only after you grant permission.
Can I share a space with someone else? Not yet. Meta says it's exploring sharing and collaboration.
The bottom line
Muse was a fast, free, handy agent that got messy the moment you trusted it with more than a few tasks. Spaces fix the mess by giving each job its own self-updating home. Voice makes creating those homes almost effortless. And free Active Memory and Active Focus make the agent behind them more dependable.
It still isn't the most powerful AI you can use, and you should be deliberate about what you connect. But if you tried Muse before and gave up because everything piled into one chat, this is the update that earns a second look. Your next personal app might take one sentence, spoken on a walk.





