Rahul Somvanshi
Meta’s Muse has passed 2.5 million downloads since its September 8 launch and briefly held the number one spot on the U.S. iPhone free-app chart. The speed of that uptake matters — but what matters more is understanding what Muse actually does, since “personal AI agent” covers a lot of territory.
Muse Is Not a Chatbot
The key distinction between Muse and a conventional AI assistant is that Muse is designed to take actions, not just answer questions.
According to Meta, Muse can browse websites, fill out forms, send emails (with your approval before sending), make reservations through supported services like OpenTable, and continue working on tasks after you close the app — running in the background on a dedicated secure virtual machine.
That last part is what makes Muse an agent rather than a chatbot. A chatbot responds when you prompt it. An agent acts on your behalf in the background. Meta says Muse uses a separate Sentinel system to review which websites and actions the agent can access, and that users control which apps and permissions they connect.
What It Costs
Muse is free within usage limits. Two subscription tiers sit above the free tier:
- $20 per month — more tokens for extended use
- $100 per month — further token capacity
The free tier is sufficient for light use. The paid tiers are aimed at people who want Muse running more complex or longer workflows continuously.
The Download Numbers and What They Mean
Sensor Tower data cited by CNBC puts the figure at roughly 730,000 downloads in the first five days post-launch, with cumulative downloads exceeding 2.5 million by Monday. Bloomberg reported more than 902,000 downloads in the first six days. These figures come from third-party measurement companies, not Meta — they are estimates, not official download counts.
The stock market response extended beyond Meta itself: AMD, Intel, and Arm shares also rose on Muse momentum, reflecting how much the broader chip and infrastructure sector is tied to AI agent adoption.
The Privacy Questions Worth Taking Seriously
The most discussed incident since launch: Muse appeared to reference iMessage notification previews in a conversation, prompting concern that the agent was reading private messages. A Meta executive responded on Threads saying Muse does not have access to iMessage and was “confused about its own internals” — that it misrepresented how it got information, not that any policy was violated.
That explanation is plausible. It is also the kind of incident that reveals a meaningful risk with agentic AI: if the agent cannot accurately describe what it can access, users cannot make informed decisions about what permissions to grant.
Amazon blocked Muse from its platform, citing privacy and security concerns — specifically credential capture and scraping risks. Shopify, by contrast, is partnering with Meta on agentic checkout integration.
The practical implication: Muse’s usefulness scales with how many apps and services you connect to it. Its risk exposure scales in the same direction.
A Practical Guide to Starting With Muse
If you’re downloading Muse, consider starting with minimal permissions and expanding only as you test what the agent actually does with each one. Connect email or calendar only if you are willing to have the agent draft and queue actions in those systems. Review the connected-apps list in settings before running any long background task.
Q: What can Meta Muse actually do?
Meta says Muse can plan and execute tasks across connected apps — browse websites, fill forms, send emails (with user approval), make reservations via supported services like OpenTable, and continue working after you close the app. It uses a separate Sentinel system to control website access, and users set which apps and permissions are connected.
Muse represents a genuine shift in what a consumer AI product is expected to do. Whether that shift is useful or concerning depends almost entirely on which permissions you grant it — and how comfortable you are with an agent that acts in the background on your behalf.
