Muse is a personal AI agent from Meta, launched in the US on September 8, 2026 on iOS, Android and muse.ai.
- This guide provides comprehensive, actionable information
- Consider your specific workflow needs when evaluating options
- Explore our curated AI Assistants tools for specific recommendations
What Muse actually is
Muse is a personal AI agent from Meta, launched in the US on September 8, 2026 on iOS, Android and muse.ai. Meta's announcement pitches it plainly: "It doesn't just answer questions, it actually does the work." You talk to it in something that looks like a text thread, give it a name and an avatar, and hand it tasks: send this email, book that trip, fill in this form, turn a saved recipe reel into a grocery list, find a cheaper bill. It keeps working after you close the app and comes back when it needs a decision.
It is not the Meta AI chatbot with a new name. It runs on Meta's Muse Spark model, and every user's Muse lives in its own cloud machine, the Muse Secure VM, with its own browser. That machine is what lets it log into sites, click through checkouts and carry on for hours without you.
What it costs
There are three tiers. The free tier is metered: you get a usage allowance and a meter showing how much of it you have used, and Meta expects most people to stay on it. Power is $20 a month and Maximum is $100 a month. Meta has not published exactly how much usage each tier includes. According to TechCrunch, Meta's longer-term plan is to take a small fee on transactions Muse completes, which is worth knowing: the business model rewards it for getting you to buy things.
What it asks for, and what it says it protects
Muse is only as useful as what you connect to it, and the list is long: email and calendar, payments, health and fitness apps, smart home, shopping, dining, music and events. You connect services one at a time and can set the level for each. Email, for example, can be read-only or allowed to send. Meta says Muse asks for your approval before it sends an email or makes a purchase, keeps a full log of what it has done and plans to do, and will forget specific things if you tell it to. You can also opt out of your conversations being used for training.
The security design is the most interesting part. Credentials live in the VM, but Meta says Muse itself cannot see your passwords or payment methods. A separate component, the Sentinel, sits between the agent and the internet: "Nothing Muse does reaches the internet unless the Sentinel approves it." Payments go through Stripe Link, with Link's purchase protections, and Shop Pay. Meta also says Muse does not share your conversations or VM data with its ad systems, and that an end-to-end encrypted Muse Confidential VM, using keys only you hold, is coming later in 2026.
These are Meta's claims, and nobody outside Meta has verified them yet. TechCrunch's launch coverage set them against Meta's record: the 2011 FTC settlement over deceiving users about privacy, the $5 billion penalty in 2019, and Cambridge Analytica. The design is a real one. Whether to trust it is the question every buyer has to answer for themselves, and it depends on how much you plan to connect.
Where it has already hit walls
On Sunday, September 20, Amazon started blocking Muse. Anyone asking Muse to shop there now gets Amazon's message that "continued access by an unauthorized AI agent violates Amazon's Conditions of Use." According to GeekWire, Amazon says Meta never told it Muse would be shopping there, and that Muse does not identify itself as an AI agent while browsing. This is not specific to Meta: Amazon has sued Perplexity over its Comet browser and moved to block shopping agents from Google and OpenAI too. If your shopping lives on Amazon, Muse cannot help with it today.
What Muse can shop is growing through partnerships instead. At Connect on September 23, Meta named Best Buy, Gap, Sephora, Walmart and Wayfair among retailers, Expedia and (soon) Instacart among services, and GitHub, Notion and Granola among work tools.
What is coming, and what is already here
Who it is actually for
For the models underneath, see Muse Spark and Meta's open-weight agent model Muse Glimmer. For what happens when an agent treats a blocked site as a puzzle, OpenAI's rogue agents is the other side of the same shift. The launch week is covered in the September 21 and September 24 briefings.
Explore curated tools related to this guide: