Muse is Meta's personal AI agent, announced on 8 September 2026, and the most useful thing to understand about it is what it is not. It books, fills in forms and chases things on your behalf. It does not transcribe your conversations, it does not produce meeting notes, and it does not hand you a record of what anyone said. If you arrived here from the AI memory wearable world, that distinction is the whole story, and it is not one Meta spells out.

Muse does Muse does not
Take actions for you: open a browser, fill out forms, negotiate on your behalf Transcribe conversations
Remember your preferences and bring them up unprompted Produce notes or summaries you can read back
Run inside an isolated virtual machine Meta calls the Muse Secure VM Publish anything about capturing audio, either way
Let you inspect its memory and tell it to forget things Give you a searchable history of what was said to you

What Meta Actually Announced

Meta describes Muse as "a secure, private personal AI agent that proactively helps people meet their goals and suggests ideas". The emphasis is on doing rather than answering. Meta's own example of its capability is that it can "open a browser, fill out forms, and negotiate on their behalf".

Two pieces of Meta terminology are worth knowing because they will appear in coverage. The Muse Secure VM is the isolated virtual machine the agent runs inside, which is how Meta separates the agent's activity and credentials from everything else. Muse Spark is the model powering it.

On availability, Meta says Muse "is rolling out in the US on iOS, Android, and muse.ai, and coming soon to AI glasses". So it is live today, in the United States, as software. The glasses support is a stated plan rather than a shipping feature.

Memory Is Not the Same as a Record

This is the part that matters most if you are comparing Muse against anything in the memory or note-taking category, and it is a genuine distinction rather than a quibble.

Meta says Muse "remembers what matters to a person, so it can make suggestions unprompted", and describes it "learning from conversations, reflecting on what matters to them". You can also "tell it to 'forget' specific things it's learned".

Read carefully, that describes a profile. Muse builds a model of your preferences, your goals and your patterns, so that next time it can act more usefully. What it does not describe is a transcript. There is nothing in Meta's material about producing a document you can open and read, search for a name, or check against your memory of a meeting.

The practical difference: Muse might learn that you prefer aisle seats and morning flights, and use that when booking. It will not tell you what your colleague actually said on Tuesday afternoon when they changed the deadline. Those are different products solving different problems, and the word memory gets used for both.

Does Muse Take Notes? No, and Meta Never Claims It Does

To be precise about the evidence: there is no transcription, note-taking or conversation-capture claim anywhere in Meta's announcement or its accompanying security material. Meta does not say Muse refrains from note-taking. It simply never mentions it, because that is not what the product is for.

That is a meaningful absence in a market where nearly every adjacent product leads with it. Plaud, Pocket, Otter, Granola and the whole AI pendant category exist to give you back a record. Muse exists to reduce the number of things you have to do. Those are not competing claims and you may well want both.

If a written record of your conversations is the thing you are shopping for, Muse is not the tool, and no amount of it being a good agent will change that. We wrote about which tools actually produce a record, and which ones only appear to, in ADHD Tool That Turns Conversations Into To-Dos.

Does Muse Listen? Meta Has Not Said

An honest gap rather than a finding. Meta's announcement and its security post contain no statement about microphones, audio capture, recording or transcription at all. Not a claim that it listens, and not a claim that it does not.

We are not going to fill that in from guesswork, and you should be wary of anyone who does. It may simply be irrelevant today because Muse is an app you type or talk to on demand. It may become relevant when the promised AI glasses support arrives, since glasses have microphones and an always-available agent on your face is a different privacy proposition from one in an app.

What Meta has published instead is about agent security: the isolated virtual machine, keeping credentials separated, and defending against prompt injection, which is the attack where hidden instructions in a web page hijack an agent that is browsing on your behalf. That is a real and serious problem for any agent that can act with your logins, and it is sensible that Meta led with it. It is a different question from what happens to audio.

What It Costs

Meta's only published statement on price is a sentence: "It's free for most of what people need, with subscription plans for people who want to do more." No figures appear on any Meta page we could reach.

Several outlets have reported two paid tiers at $20 a month and $100 a month. That reporting is consistent across TechCrunch and Axios, but it is third-party rather than confirmed by Meta, and we are labelling it that way rather than presenting it as a price list. If the number matters to your decision, check it at the point of signup.

One practical caution. Meta's announcement points to muse.ai as the web destination. At the time of writing, that domain still served an unrelated video search product with its own subscription tiers. Do not mistake those prices for Muse's.

Who Muse Is Actually For

It fits if your problem is the volume of small administrative tasks: bookings, forms, follow-ups, comparing options, chasing things you started and abandoned. An agent that can drive a browser genuinely removes work, and that is a category worth taking seriously.

It does not fit if your problem is losing information. If what goes missing is what people told you, what you agreed to, or the detail from an appointment, an agent that builds a preference profile does not touch that. You would be buying a very capable assistant for a job it is not doing.

It also does not fit if you are outside the United States, at least for now.

That second case is the one we build for. Legend One is a voice-first AI memory pendant made for people with ADHD, worn at chest height with three microphones and speaker separation, always on with no button. It transcribes speech to text instantly across more than 150 languages, and no audio file is written at any point. Transcription happens on the device and the text is stored locally, then syncs to your phone when you are back in Bluetooth range. A Halo light shows when it is active. It is $499 USD including the first year of membership, renewing at $499 a year, which is $1,497 over three years, with a 30-day money-back guarantee.

It is also worth saying plainly that Legend does the opposite job to Muse. It gives you the record and it does not book your flights. If your difficulty is administrative rather than mnemonic, a $499 pendant is the wrong purchase and an agent is the right one.

The Takeaway

Muse is an agent, not a notebook, and the distinction is easy to miss because both get marketed with the word memory.

Try Muse if you are in the US and want something that can act on your behalf across the web. Meta says the useful tier is free, so the cost of finding out is low. Do not expect a transcript, because there is no indication Muse produces one, and nothing in Meta's material suggests that is coming. Watch the glasses, because an agent on a device with microphones raises questions Meta has not yet had to answer. Look elsewhere for a record if what you actually need is what was said, whether that is a meeting tool, a recorder, or a wearable.

FAQ

What is Meta's Muse? A personal AI agent announced on 8 September 2026 that takes actions on your behalf. Meta describes it as able to open a browser, fill out forms and negotiate for you, and says it is rolling out in the US on iOS, Android and the web, with AI glasses support coming.

Does Muse take notes or transcribe conversations? No. There is no transcription, note-taking or conversation-capture claim in Meta's announcement or its security material. Muse has memory in the sense of a preference profile it builds about you, which is different from a readable record of what was said.

How much does Muse cost? Meta says only that it is "free for most of what people need, with subscription plans for people who want to do more", and has published no figures. Several outlets have reported tiers at $20 and $100 a month, but that is third-party reporting rather than a Meta price list.

Does Muse listen to you? Meta has not said. Its published material contains no statement about microphones, audio capture or recording, in either direction. The question may become more important when the promised AI glasses support arrives.

Sources: Meta Newsroom, Introducing Muse; Meta Research, Security and safety for AI agents; pricing reported by TechCrunch and Axios, not confirmed by Meta. All Meta quotations are from Meta's own pages, verified 11 September 2026. Prices are USD.