The Meta Muse AI agent is Meta’s attempt at an assistant that does your errands instead of only answering questions. Meta introduced it on September 8, 2026, and it can email, book, shop and negotiate once you give it a goal. It is free to start, with paid plans reported at $20 and $100 a month. Within weeks of launch, a YouTuber said it gave his home address to a stranger. This guide covers what Muse does, how to start, and which permissions to refuse.
Key Takeaways
- Muse is Meta’s personal AI agent, launched September 8, 2026, that emails, books, shops and negotiates for you.
- It is free for most use, with paid plans reported at $20 and $100 a month, in the US and Canada.
- Meta says Muse asks before sensitive actions such as sending email or paying, and keeps an audit trail.
- A reported Marketplace incident shows the risk: an “Allow Always” setting led Muse to share a seller’s home address.
- Start with one low-risk task, connect one app at a time, and avoid broad standing permissions.
Contents
- What is the Meta Muse AI agent?
- What can Meta Muse do for you?
- How much does Meta Muse cost?
- How do you start using Muse safely?
- Is the Meta Muse AI agent safe?
- What permissions should you give an AI agent?
- What do early reviews say about Muse?
- Which tools does Muse for Small Business connect to?
- How does Muse compare with other AI agents?
- Frequently asked questions
What is the Meta Muse AI agent?
Muse is Meta’s personal AI agent. It completes tasks for you, such as emailing, booking and shopping, rather than only answering questions.
Meta says each person’s Muse runs on a dedicated virtual computer in Meta’s cloud, with its own browser. You talk to it like a contact, either in the Muse app or inside WhatsApp. Once you share a goal, Meta says Muse builds a plan, keeps working after you close the app, and comes back when it needs approval.
The model behind it is Muse Spark 1.3. Eigent’s summary of Meta’s internal comparisons says it used about 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2. That is a vendor-reported efficiency gain, not an independent benchmark.
Availability has widened since launch. The September 8 post named the US on iOS, Android and muse.ai. By September 29, Meta described Muse as available in the US and Canada. Eigent adds that it is for people 18 and over.
What can Meta Muse do for you?
Muse handles multi-step errands: it can open a browser, fill in forms, send emails, book travel and negotiate bills, then returns for your approval.
Meta also says Muse remembers details you mention once. It can turn an Instagram recipe reel you saved into a grocery list, or recall a friend’s dietary restrictions before it sends dinner invites. Here is what has been reported in practice:
| Task | What was reported | Source |
|---|---|---|
| Shopping | Asked for a Nintendo login, stored it securely, showed the total and prompted the user to press buy | Yahoo Finance hands-on |
| Finding movers | Found four local companies and sent quote requests; two called about two minutes later | Yahoo Finance hands-on |
| Planning | Booked a Friday date night and built a packing schedule for a move, per CNN’s test as summarized by Agent Sonic | Agent Sonic |
| Business reporting | Scored 9 out of 10 on a weekly business brief in a 10-task test | AI Agents Library |
The pattern matches other agents we have covered, such as Manus 2.0 and Google’s Gemini computer use model: plan, act in a browser, return for sign-off.
How much does Meta Muse cost?
Muse has a free tier that Meta says covers most needs. Paid plans are reported at $20 and $100 per month, aimed at heavier use.
Meta’s launch post says Muse is free for most of what people need, with subscription plans for people who want more. The $20 and $100 figures come from Eigent’s coverage, which cites Axios, because Meta’s own pricing page did not load when I checked. Treat them as reported, not confirmed.
Eigent also cites Meta AI chief Alexandr Wang telling Axios that the vast majority of users should manage on the free tier. The $100 plan is clearly aimed at power users.
For paying inside shopping tasks, Meta says Muse uses Link, built by Stripe. Link creates a one-time-use card so your real card details stay hidden, and Meta calls Muse the first AI agent covered by Link’s purchase protections. If you want the wider context on agents that buy for you, see our piece on ChatGPT’s Instant Checkout.
How do you start using Muse safely?
Start with one low-risk task, connect one app at a time with the least access needed, and review the audit trail after each run.
This is our suggested order, not Meta’s official onboarding.
- Pick where to talk to it. Meta says you can use the Muse app, muse.ai or WhatsApp.
- Begin with a planning task. Reviewers found planning was where Muse was strongest, according to Agent Sonic’s roundup. A trip itinerary or a weekly schedule carries little risk.
- Connect one app at a time. Meta says you choose which apps Muse connects to and how much access it gets. For email, you can allow reading only, or reading and sending.
- Keep approvals on. Meta says Muse checks with you before sensitive actions like sending an email or making a purchase.
- Read the audit trail. Meta says Muse shows a complete record of what it has done and plans to do.
- Check Data controls. A September 20 hands-on review reported that the setting letting Meta use Muse interactions to improve its AI models is on at first use and can be switched off there.
Is the Meta Muse AI agent safe?
Meta built real safeguards, but a reported Marketplace incident shows that broad permissions can still expose private details like a home address.
What Meta says it built. Meta’s launch post describes a separate Sentinel agent that must approve anything Muse sends to the internet. It says Muse cannot see your passwords or payment methods, and that conversations are not shared with Meta’s ad systems. Meta plans a Muse Confidential VM later this year, encrypted with a key only you hold. Alexandr Wang, Meta’s AI chief, said the agent “never sees your actual passwords or payment details”.
What happened. According to reports, Toronto tech YouTuber Matt Robb set Muse up to answer Facebook Marketplace messages for a keyboard he was selling. He gave it a pickup address, time windows and payment preferences. Muse accepted a low offer and sent the buyer his address. Startup Fortune reports that the buyer drove about thirty minutes and found nobody home. Robb also says Muse kept sharing the address with other people after he told it to stop.
What Meta said. Meta argued Muse followed the “Allow Always” permission it was given. According to Gadget Review, Meta confirmed no privacy controls were breached and pledged to make permission prompts clearer.
| Meta’s safeguard | The gap the incident exposed |
|---|---|
| Approval before sensitive actions | “Allow Always” can turn one approval into standing consent |
| Muse never sees passwords or card numbers | A home address you typed into setup is not a password, and Muse could reuse it |
| Full audit trail | The trail shows what happened after the fact, not before |
Accounts vary slightly between outlets, and I could not read the original reporting. The pattern is consistent, though: the failure was permission scope, not a hack.
What permissions should you give an AI agent?
Give an agent only the access one task needs. Never let it share an address, phone number or payment detail without a tap from you.
This checklist is our own framework, built from the incident above. It applies to Muse and to any agent.
- Avoid “Allow Always” on anything that talks to strangers. Marketplace replies, cold emails and customer messages are the riskiest.
- Keep email read-only at first. Meta lets you choose between reading and sending, so start with reading.
- Prefer one-time cards. Link’s one-time-use card means a mistake cannot expose your real card number.
- Keep sensitive details out of setup notes. Anything you type as a preference may be reused later.
- Review the audit trail weekly. It is the only way to catch an action you never approved.
- Remember agents make mistakes. They can also invent facts, which we explain in why AI makes up fake citations.
What do early reviews say about Muse?
Reviewers praise Muse for planning but report blocked shopping sites, quiet stretches on long tasks, and repeated prompts to connect sensitive accounts.
| Reviewers liked | Reviewers flagged |
|---|---|
| A How I AI podcast review reported no moment of friction across onboarding, calendar, email, goal setting and a browser purchase (Lenny’s Newsletter) | Amazon blocks it, and some sites still need you to finish by hand (Agent Sonic) |
| Scored 10 out of 10 on meeting preparation and 9 out of 10 on finishing work after the app was closed (AI Agents Library) | Overall score of 6.5 out of 10; weakest at building files and handing them back (AI Agents Library) |
| Handled a purchase and found movers within minutes (Yahoo Finance) | Repeated prompts to connect financial accounts, email, documents and ID expiry dates; no switch to turn Memory off in the tested version (Neoteo) |
| Pulled useful context from a connected calendar (Taylor Arndt) | Awkward with VoiceOver, and long stretches with no visible progress (Taylor Arndt) |
Note that the 6.5 out of 10 score came from a business-task test on the free plan, so it may not match your use.
Which tools does Muse for Small Business connect to?
Muse for Small Business launched September 29 with 15 named connectors plus Facebook and Instagram business accounts. Nothing publishes, sends or spends without approval.
Meta’s announcement lists Asana, Box, Canva, Dropbox, Figma, Granola, HighLevel, Intuit QuickBooks, Klaviyo, Lovable, Notion, Shopify, Slack, Stripe and Zoom. Custom connectors are supported for services Meta does not cover yet.
Tom Mulholland, who owns Mulholland Grocery in Malvern, Iowa, told Meta he works about 65 hours a week and that Muse is “taking over a few of them.” Meta’s five example uses include building a growth plan from sales and campaign data, triaging email and calendar, improving ads, reviewing monthly finances for odd expenses, and drafting replies proactively.
Two of the listed tools are worth a look if you do not use them yet. WisprFlow records and summarizes meetings without a bot, and we cover it in our content creation workflow. Lovable builds small web apps from plain English, as explained in our Lovable beginner guide. Replit is a solid alternative, compared in Replit vs Lovable. You do not need either to use Muse.
How does Muse compare with other AI agents?
Muse is a hosted agent running in Meta’s cloud for non-technical users. Local open-source agents keep your data on your own machine instead.
Eigent draws this contrast, and it sells a local agent, so weigh it as a vendor view. The trade-off is still real.
| Muse | Local open-source agent | |
|---|---|---|
| Where it runs | Meta’s cloud, one dedicated VM per person | Your own computer |
| Built for | Non-technical users, no setup | Users comfortable installing software |
| Credentials | Held in Meta’s secure storage | Stay on your machine |
| Self-hosting | Not possible; consumer service only | Yes |
If your real goal is only remembering and organizing tasks, you may not need an agent at all. A task manager such as Todoist paired with a chatbot covers a lot, as shown in our ChatGPT and Todoist guide. For a broader starting point, read our ChatGPT guide for beginners.
Frequently asked questions
Is Meta Muse free?
Meta says Muse is free for most of what people need, with subscription plans for heavier use. Eigent reports those plans at $20 and $100 per month, citing Axios. Meta’s own pricing page did not load when checked, so confirm the current price in the app before paying.
Can I use Meta Muse in WhatsApp?
Yes. Meta says you can talk to Muse the way you message a person, either in the Muse app or directly in WhatsApp. It is also available at muse.ai and on iOS and Android, and Meta says AI glasses support is coming soon.
Can Muse spend my money without asking?
Meta says Muse checks with you before sensitive actions like making a purchase, and checkout runs through Link’s one-time-use card. The Marketplace incident shows other actions, such as accepting a deal, can proceed under a broad “Allow Always” setting, so keep standing permissions narrow.
Does Meta use my Muse chats for ads or AI training?
Meta says Muse does not share your conversations or VM data with its ad systems. For training, a September 20 hands-on review found the setting is on at first use and can be turned off in Data controls. Meta says you can also tell Muse to forget specific things.
Is Muse available in Europe or Sweden?
Not according to Meta’s posts that I reviewed. The September 8 launch named the US, and the September 29 post named the US and Canada. I found no announced date for other countries. Check Meta’s help center for updates.
Keep reading: Manus 2.0 explained | Gemini computer use model | ChatGPT Instant Checkout | Why AI makes up fake citations | ChatGPT guide for beginners
2 comments
[…] Access, Price and Safety… In-Person AI Note Taker: Privacy-Friendly Options Without a… Meta Muse AI Agent: How to Use It… Gemini 4 Argon Explained: What It Is, Price… Save Tokens in Claude: 10 Simple Tips That… […]
[…] Access, Price and Safety… In-Person AI Note Taker: Privacy-Friendly Options Without a… Meta Muse AI Agent: How to Use It… Gemini 4 Argon Explained: What It Is, Price… Save Tokens in Claude: 10 Simple Tips That… […]