Why Is Muse So Much Better Than the Competition?

If you had to bet on who would build the assistant that knows you the best, Google would be a safe pick. Gemini can see your email, calendar, photos, web searches, maps, every video you’ve watched and a lot more. Yet Google’s own personal assistant, Gemini Spark, has barely been mentioned since it launched in May, and this month Google began folding it into the regular Gemini app and dropping the Spark name.

But the best personal AI assistant came from Meta instead. Muse launched in September and within days it felt more personal than any other AI assistant, including ChatGPT/Dots and Claude that we use every day. So how did Muse become such a great personal assistant almost overnight?

Muse’s heartbeat

In our testing, Muse runs a check-in every 30 minutes, known as a heartbeat, working through a checklist it writes for itself and looking for anything that needs doing. It also has a look over your personal context during every heartbeat. That’s 48 check-ins a day, even on days you never open the app. TIME reported that Muse also rewrites its notes about you every hour and reviews each day’s conversations every night, so by morning it has already documented what changed in your life.

Those check-ins explain a lot of why Muse feels like it knows you so well. A great human assistant doesn’t need a briefing, because they check in all day and already know what you’re up to. Meta borrowed the 30-minute heartbeat from OpenClaw, a free assistant programmers set up for themselves, but Meta is the first to put one in a free app for everyday people. OpenAI’s Dots and Google’s Gemini Spark only do that on paid plans.

Trained to remember you

The check-ins keep Muse current, and the model decides what to do with what it knows about you. When Meta launched its newest model, Muse Wonder, Mark Zuckerberg told the Financial Times where Meta is heading: “Everyone else is focused on more knowledge work, automation-type cases, and we’re much more focused on personal empowerment.”

The other labs trained their models focusing on code and math, where a model training run can verify millions of answers automatically. Meta trained for something harder to score, remembering what matters to one particular person, and wrote its own test for it called MemoryBench, which measures how well a model remembers you and keeps learning.

Meta says its new model Wonder beats leading models from OpenAI and Anthropic on knowing the customer. “We put a tremendous amount of effort into making sure your model can remember what’s important to you,” Meta research scientist Kavya Kopparapu said in the launch video. The new model will power Muse soon, and it’s safe to say the same training techniques are already used in today’s version.

Every Muse learns from the others

Each person’s Muse runs as its own copy on Meta’s computers, and internal instructions reported by TIME say those copies “teach each other through shared lessons.” Each copy strips out names before passing on lessons about what customers prefer, and Meta says the lessons go into the product as a whole, not straight from one Muse to another.

Other labs train on their users’ chats too, so our guess is that Meta’s edge comes from the way it trains on this feedback. With millions of downloads, every assistant checking in every half hour and lessons about customer preferences flowing back to the core system, Meta is building a record of what people need like no lab has done before. That record is built from its users’ own lives, and it’s a training method the other labs don’t seem to have tried.

Thirty minutes today, real time soon

Every personal AI assistant will copy the check-in, if they can justify the economics, and in AI a lead lasts a few weeks. Expect the time between heartbeats to drop from half an hour to a few minutes and then to seconds, with each check-in looking for tasks and pulling in whatever just happened to you.

The heartbeat, the training and the shared lessons all aim at one thing, knowing you better than any other assistant does, and that is what puts Muse ahead of the rest. How much of your life your context library covers, and how fresh it is, will decide how good any assistant gets.