Fetching from the wire…
Public story · 2026-09-09 · high
Every Muse user gets a dedicated cloud machine so the agent can act inside apps a person already has open, with a free tier and two paid plans.
Why now: Meta shipped Muse to US users 18 and up on September 8.
Meta released Muse on September 8, a personal AI agent that runs inside its own secure virtual machine for every user. It's available now through a standalone app or through WhatsApp, to US users 18 and older.
Anyone running a self-built agent sandbox has had to provision that isolation themselves. Meta just put a $20-a-month price on it for consumers.
The infrastructure choice is what stands out. Each person gets a Muse Secure VM, a dedicated machine with its own browser, so the agent can act across the apps someone already uses. That's the shape independent agent-harness builders have been wiring together by hand for the past year. Give the agent a real browser in an isolated box, and it can click, type, and navigate like a person would. Meta made that the default architecture for a mainstream product instead of a setup only a developer would attempt.
Pricing splits three ways. A free tier is reported at up to 100 million tokens a week, Power runs $20 a month, and Maximum is $100 a month. That's a consumer price tag on per-user VM compute most people would otherwise have to run themselves.
On privacy, Meta says Muse conversations and VM data don't feed its ad systems. It's also promising a Muse Confidential VM later in 2026, encrypted with a key only the user holds, one Meta can't open. Meta's announcement doesn't say what month that confidential tier ships, or what happens to VM data before then beyond the current ad-system carve-out.
Each link below shares sources, entities, or timing with this story.
Muse Image, from Alexandr Wang's lab, generates and edits from text, writes and runs code to produce accurate plots and QR codes, and searches the web to ground outputs. It launched free in Meta AI, WhatsApp, and Instagram Stories, and immediately drew pushback for training on...
For about a year, "run your agent locally" meant accepting a model that couldn't reliably call a tool twice in a row. That excuse is gone. Meta Superintelligence Labs published Muse Glimmer today: a 29.6B dense causal transformer, 52 layers, 6,656 hidden dim, with a ~1.8B ViT-...
Launched August 8 as the new Quality Mode at grok.com/imagine and in the Grok mobile apps, pitching precision editing, crisp text rendering, and improved factuality, with API access promised but not shipped (The Decoder). On the August 7 Arena leaderboards the faster "low" var...
Four chips detailed: MTIA 300 (in production), 400 (lab testing), 450 and 500 (GenAI inference, 2027). From 300 to 500: HBM bandwidth up 4.5x, compute FLOPS up 25x. New chips every six months. The most aggressive custom silicon roadmap from any hyperscaler, reducing NVIDIA/AMD...
Reads, edits, and executes actions on local files and applications directly on the user's machine. Positions Meta against OpenClaw and Claude Cowork in local agent runtime. No prior announcement preceded the release. OneNewsPage ---
$1.25 input / $4.25 output per million tokens for the standard tier. $0.10 / $0.20 for the Contributor tier, where Meta trains on your usage and feedback. That's roughly 12x on input, 21x on output, and it's the clearest number anyone has published on what your proprietary sou...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.