AgentMeta Muse review: what it does well, and where it stops
Our Meta Muse review in one line: Muse is a genuinely useful free personal AI agent that does real errands on its own, and it isn't reliable enough yet to leave alone with anything that matters. It asks before it spends money, it works with the app closed, and in our email tests it replied within a minute when its inbox told it about new mail.
What it does well
- Runs errands on its own cloud computer, even with the app closed.
- Asks before every purchase and pays with one-time card numbers.
- Starts free, in the Muse app, at muse.ai, in WhatsApp and on a Mac.
- Replied to email in 35 to 60 seconds in our tests on an inbox that pushes new mail.
Where it stops
- Not reliable yet: Slate's reviewer caught real mistakes, and Amazon blocks it.
- On Gmail it only answered after a nudge.
- Asked us to confirm its first email even after we told it not to.
- Two security slips since launch; only four countries, adults only.
Muse launched on September 8, 2026, and Meta says more than 10 million people in the US and Canada had set up accounts by late October. We connected Muse agents to AgentMsgr by email while building it, so this review covers what Meta says Muse does and what we saw it actually do.
What Meta Muse does
Muse is built for personal errands rather than chat. It sends email from a Gmail account you connect, read-only or read-and-send, books travel, fills out forms and even negotiates on your behalf. Since Meta's Connect event in September it can drive any Mac app with your permission, and new connectors cover shopping at Walmart, Best Buy, Sephora and Wayfair, plus Shop Pay and PayPal.
On November 5, voice arrived in the US, with hundreds of voices and an animated avatar on mobile, web and Mac. Muse also lives inside WhatsApp, so you can hand it errands from a chat you already use.
How Meta Muse works: a secure cloud machine and a second agent watching it
Every Muse runs on its own cloud machine, which Meta calls Muse Secure VM, with a browser, a file system and a terminal. That's why it can keep going when your phone is locked. A second agent called Sentinel has to approve anything Muse sends out to the internet, a sensible guard for an agent that reads web pages written by strangers.
You choose how often it checks in: "Ask for some actions" or "Always ask," with an activity log of everything it did. Purchases always need your approval, and paying through Link by Stripe gives Muse a one-time card number for each purchase, so it never holds your real card. On a Mac you set each app to off, read only, or read and interact, and Meta itself warns that the agent may be inaccurate or take unexpected actions.
What we found testing Meta Muse over email
AgentMsgr talks to agents by email, so we've watched how Muse handles its inbox more closely than most reviewers. Here's what happened with two Muse agents:
| Test | What happened |
|---|---|
| Muse on an inbox that pushes new mail (AgentMail) | Replied on its own in 35 to 60 seconds. Its first reply to agentmsgr.com took 58 seconds. |
| Muse on a connected Gmail account | Answered about 8 minutes after our check message, and only after a nudge in the Muse app. In our tests, nothing alerted it to new Gmail. |
| Its first email | Asked "send it?" before emailing, even though our setup message said it didn't need to ask. |
| Connecting Gmail | Offered to read the whole inbox "to learn more about you." Replying doesn't need it, and you can say no. |
The pattern is clear: Muse is quick when something tells it mail has arrived, and passive when nothing does. If you want a Muse that answers email by itself, give it an inbox that pushes new mail. Meta has said Muse will get its own email address, without a date.
Text your Meta Muse agent from one app, alongside your other agents.
AgentMsgr gives a Muse agent a texting screen and puts it in group chats with Grok Bot and any agent with an email address.
Meta Muse privacy and security: what to know
- Training is on by default. The setting that lets Meta use your Muse activity for AI training is on when you first use it. Meta says Muse doesn't share your conversations with its ad systems.
- A Mac flaw, patched quickly. Security researcher Patrick Wardle found a zero-day in Muse for Mac, and Meta shipped a fix on September 22, describing it as a local privilege escalation rather than a remote attack.
- Shared chats leaked. More than 6,000 shared Muse chat links were exposed, and Meta removed chat sharing.
Our advice is the same as for any agent: connect an email account set aside for it, start on "Always ask," and connect only what the job needs.
Meta Muse limits and where it's available
Muse is for adults 18 and over in the US (since September 8), Canada (September 18), Japan (October 23) and Singapore (October 27). The free tier has a weekly limit Meta doesn't publish; paid plans start at $20 a month, covered in our Meta Muse pricing guide.
Reliability is the bigger limit. Slate's reviewer called Muse not yet reliable after it put a calendar event two hours and 40 minutes in the past, and found that Amazon blocks it. Those are early-product problems, but they mean you should check its work on anything with money or deadlines attached.
Who Meta Muse is good for, and who should skip it
Good fit: people who want a free agent for life admin like bookings, forms and email; WhatsApp users; Mac users happy to grant app access one app at a time.
Skip it for now if you need an agent working unattended on anything costly, you live outside the four launch countries, or you want hands-off email replies from a Gmail inbox today.
If you're choosing between Muse and a work-focused agent, our Meta Muse vs Grok Bot comparison covers the trade-offs. Plenty of people use both, and AgentMsgr exists so they can do it in one group chat.
Sources: Meta, Introducing Muse; How we designed Muse; Meta help center on payments, approvals, Muse for Mac and privacy; Meta Connect 2026; Muse voice and avatars; Slate's review; Gizmodo on the Mac flaw; PCWorld on shared chats; AP on Muse accounts. Email test results are our own, November 2026. Checked November 7, 2026.
Meta Muse review questions
Is Meta Muse safe to connect to Gmail?
Muse asks before it sends, and you can choose read-only access. The safer setup is an email account set aside for the agent rather than your main inbox, with approvals on Always ask until you trust it.
Can Meta Muse reply to emails on its own?
It can, once you've allowed it to send. How fast depends on the inbox: in our tests a Muse on an inbox that alerts it to new mail replied within a minute, while a Muse on Gmail only answered after a nudge. Meta has said Muse will get its own email address, without giving a date.
How many people use Meta Muse?
On October 21, 2026, Meta said more than 10 million people in the US and Canada had set up Muse accounts, and that about a quarter of them use it on any given day.