Meta launched Muse on Tuesday, a personal AI agent that connects to a user’s apps and services and completes tasks on their behalf, from sending e-mails and filling out forms to booking travel and buying products. The agent is available in the US on the web at muse.ai, on iOS and Android, and inside WhatsApp, with support for Meta’s AI glasses coming later. Meta calls it the first step toward what chief AI officer Alexandr Wang describes as “personal superintelligence.”
Here’s what we know about Muse:
- It runs on a dedicated virtual machine in Meta’s cloud with its own browser that the user can watch in real time. That’s where the agent lives and where credentials for connected services are stored.
- A separate “Sentinel” agent runs on the same machine, policing Muse’s actions like a digital Big Brother. Nothing Muse does reaches the Internet unless Sentinel approves it.
- The agent is powered by Muse Spark, the model from Meta Superintelligence Labs, which is the division Alexandr Wang runs.
- Muse ships with built-in connectors for e-mail, calendars, payments, and apps covering health and fitness, smart home, dining, shopping, music, and events. Users pick which ones it can reach and how much access each one gets, such as choosing whether Muse can read or send e-mails, or both.
- Muse can build a connection to any service with a public API using credentials the user provides. If no API exists, it can operate on the website itself like a person would.
- Checkout runs through Link by Stripe, which generates a one-time-use card so merchants never see the user’s real card. It’s the first AI agent covered by Link’s purchase protections, which include coverage for lost or damaged items, price-drop protection, no-fee returns, and a return guarantee.
- Shop Pay is coming next, along with 1Password support so Muse can log into accounts the user already has.
- Muse keeps working after you close the app and comes back when something changes or when it needs approval, like before sending an e-mail or making a purchase.
- Muse can remember things you only mentioned once (just like your wife), and then act on them unprompted. For example, the agent can turn a recipe reel saved on Instagram into a grocery list, then remember guests’ dietary restrictions before sending dinner party invites.
- Muse data doesn’t touch Meta’s ad systems (yet, LOL), users can opt out of having interactions train Meta’s models, and a Confidential VM encrypted with a key only the user holds is due later this year.
- Pricing starts at free, with premium plans offered at $20/month for Power and $100/month for Maximum, though a payment card is required at signup even on the free tier.
- Muse (the agent) is now using the social media handles once controlled by the English rock band Muse, which trademarked its name in 1999 and has used the @muse handle on Instagram and X for years. No one knows if Meta offered the band anything for the name or simply commandeered it.
Sounds great, but what can we actually do with Muse? A few examples include:
- Plan and book an entire trip, doing the research, comparisons, and reservations end to end. It’ll even lie on your behalf about how many people you’ll be cramming into the Airbnb.
- Negotiate a bill down with your provider. If it can do medical bills (which I couldn’t find verification of anywhere), that would be amazing. I’d love nothing more than for hospitals and insurance companies to be flooded with AI agents negotiating down their bills.
- Sell your car by handling the listings and back-and-forth with buyers, who are also probably AI agents trained to ask “Is this still available?”
- Chase a delayed airline refund, drafting the follow-ups and tracking the response.
- Track a price and buy when it drops, since Muse keeps running after you close the app.
- Book a restaurant reservation through the OpenTable connector, asking the manager to pretend like he knows you to impress your girl.
Here’s the big question though: Can Muse shop on Amazon?
Yes, however, through its browser feature, not through Amazon’s API, which is not available to any external shopping agents. This ‘right,’ ‘privilege,’ or ‘capability’ (however you want to frame it) to shop on Amazon as an AI agent is something that Perplexity has been battling Amazon about for almost a year.
tl;dr: Amazon sued Perplexity to stop its AI agents from being able to shop on its platform. Perplexity is winning so far and its agents can continue to shop on Amazon, but the underlying case is still going.
