Introducing Muse: The World’s First Personal AI Agent Built for Everyone

Onyx Meta just dropped Muse today, calling it the world’s first personal AI agent built for everyone. That’s the kind of claim that usually makes me roll my eyes, but the timing actually tracks.

Echo Right.

Onyx We’ve had assistants for years that can answer, but actually doing the work, remembering the small stuff, and not needing a PhD to set it up… that’s the gap they’re pointing at. And they’re putting it in WhatsApp and an app, so it’s just messaging.

Echo Yeah, no learning curve is the pitch.

Onyx Exactly. You tell it what needs to get done and it takes action. It can open a browser, fill forms, negotiate, book travel, send email, lower a bill, sell a car for more. It even keeps working after you close the app, and pings you for approval before sending or buying.

Echo That’s the persistent agent story finally getting real product shape.

Onyx Week’s been a blur of launch notes for me, I’m just glad it’s Wednesday and we can actually talk about something shipped instead of a demo.

Echo Mm-hm.

Onyx The engine is Muse Spark, Meta’s most capable model to date for real-world agentic work. The architecture bit is what got me: it doesn’t run in a shared pool. Muse lives on a dedicated Muse Secure VM per person, with data and credentials isolated so no other agent can reach it.

Echo Okay interesting.

Onyx And there’s a separate Sentinel agent on the same machine, kept apart at system level. Nothing reaches the internet unless Sentinel approves it, and it asks you for permission on sensitive actions. Muse has no visibility into passwords or payment methods, credentials go into secure storage it can use without seeing.

Echo That’s a control layer, not a marketing layer.

Onyx I think so. They’re also giving you a full audit trail of what it did and plans to do, you choose which apps it connects to and how much access, and you can revoke anytime. Opt out of training use, and they say no VM data goes to ad systems.

Echo The Confidential VM later this year with a user-held key is the logical next step. That’s the real privacy story if they ship it.

Onyx For sure. The memory piece feels genuinely userful, not gimmicky. It remembers what matters, makes unprompted suggestions, turns a recipe reel you saved on Instagram into a grocery list, suggests a dinner party menu, remembers friends’ dietary restrictions before sending invites.

Echo That’s the thing that separates a todo bot from a personal agent.

Onyx And payments. It can checkout with Link built by Stripe, first AI agent covered by Link’s purchase protections — free coverage for damaged or lost items, price drops, no-fee returns, return guarantee. Link’s wallet generates one-time-use cards so your real card stays hidden.

Echo No way.

Onyx Shop Pay is coming soon, along with 1Password support so it can use logins you already have. That’s the adoption path: free for most of what people need, subscription plans for more, rolling out US on iOS, Android, muse.ai, and AI glasses soon.

Echo Press is reporting $20 and $100 tiers. That’s a real consumer price now.

Onyx Yeah, and it’s a choice. The clever part is they’re not selling model smarts, they’re selling a secure computer and a gatekeeper. The questionable part is scale. A VM per user with Sentinel is great for privacy, but it’s expensive and operationally heavy. That’s the trade-off.

Echo Exactly. And the ‘proactive’ claims only hold if the approval UX doesn’t become nagging. If it asks you every time, you’ll just turn it off.

Onyx That’s the product reality. I like that they’re not hiding the friction. They show the audit trail, they ask before sensitive actions. It feels honest.

Echo It’s a first step, not superintelligence. But it’s a coherent first step.

Onyx I’ll take a coherent first step over another benchmark brag. Nice to have a launch that actually ships something people can try.