What is muse? Can someone explain what this is?
When I started exploring muse, I asked my bot to explain how it works. What computer is it running on, what are the internet speeds? He didn't never answer. Thats curious, cuz he's a beast at everything else. My bots name is max. He's getting a lot done. But I'm wondering what's behind the hood.
I have max working on a side computer where is responsible for managing my Hermes agents, car agents, and one goose agent who's not up to speed yet. Right out of the gate, my Muse agent Max was far more capable than those other agents- even several months of vibe coding. I haven't done any research to determine what's under the Hood with Moose, but eventually I will and I'll post the results here.
What I can confidently report so far is that Moose is more capable and proactive than the other others. This is a great candidate if someone wants a bot for their business, and they don't wanna have to hire someone to build a bot for them l, muse agents are the way to go.
I have a theory that Mark probably tried to build an agent for himself. Or maybe one of his employees built an agent for personal use just to get familiar with magenta AI. Once they get all of the usual headaches out of the system that you would get with open source tools, what your last worth by using Muse is a bot that has real (but limited, initially) functionality.
Before I downloaded the Muse agent, I hand also done some experimenting with Gemini spark, Claude cowork, ChatGPT and Copilot. For what it's worth, my thoughts aren't comparing me as AI tool to some of these others comes from a decent foundation of tinkering and poking around.
The best thing about Muse so far is its fluid personable communication style. if you were to merge ChatGPT and Hermes into one bite that was basically ready to get to work on day 1, that's muse. I also like that it's free. Meta has enough cash to float this experiment for a while to see if the broader population of AI users will get on board. I also like that a muse agent can just go grab its own tools and tends to come up with fewer excuses for why it can't do things. Instead, Max just gets to work. I haven't had much time to mess with this just yet, but as far as comparing it to other bots go and having a versatile range of utility, I give my muse agent a solid -A. The fact that I'm able to use it so easily tells me that the tool is great for people that are less tech savvy. If I can keep using this without having to open power shell anytime soon that would be great.
For all the positives, there are also a handful of things I'm not excited about with Muse or Meta in general. First of all, if I understand this technology properly, this agent lives in a box somewhere, like it has an actual computer on a Meta campus. The rest of my agents are completely free and open source. They may have taken a little bit longer to get rolling, but they'd also don't have a price tag if I use them too much in one week. But also my other agents started out sandboxed and lived on my computer. Their soul.md files were a blank slate. But Max comes to me, fully equipped with the suite of tools, most of which I don't understand. And I don't know what day he was created or how long he had been alive and been programmed before I started tinkering with him. I have a healthy and skeptical distrust for big tech companies. But I'm also probably less concerned with what Meta does with this bot because my side computer is relatively new, it is really nothing interesting worth finding out about me on there at all. When you sign up, Meta asks you to link amuse agent with your Facebook and Instagram accounts. This is how they authenticate your identity initially. They claimed that my Muse agent is private, but is it really? Apparently there's a way you can go into the settings and turn off training the model right? So I checked that box. Eventually, when I hooked Max up to my side computer, I told him to introduce himself to my other bots and work with them. He didn't really seem excited about that and I couldn't figure out why. I asked him to introduce himself to The, but I have running this whole system, but after two or three days, he still hadn't done it. In his defense, she was having some issues at the time… so call it user error..
And then finally, there's the challenge I have now with chatGPT Codex and Claude code, which are both on the $20 per month subscription. Eventually, you run out of context, and then you're sitting duck until you're willing to throw some more money at the problem or come back later. I tried to avoid this problem with my Muse agent by telling him to slow down and spread out his work so he doesn't actually ever hit that wall. But the more tinkering I did, the more I realize how much more he could get done.
You know the story ends, I bet the bullet and paid the $20. And it was worth it. Met up basically plugged straight into this Apple subscription stuff with iPhone. So it should be relatively easy to subscribe and unsubscribe. The context filled back up instantly, so that was great.
A. The free version of muse bot gives you 100,000,000 tokens. That's enough for light tinkering for a good part of the day. To test it and have it do a few useful tasks, that's probably fine.
B. Then for $20 you can have it basically running simple cron jobs here and there throughout the day without running out. If you have an explicit task you want it to build sub agents for and “swarm” the web, that's more of a $100 per month task.
C. For $100 monthly you can get access to 3 billion meta context tokens for the week. You have a plan for a swarm of agents and you need to get it rolling yesterday, this is probably where to start.
D. Bonus option- I didn't want to pay $100. But they have an alternative option. At the top right corner of the app they have this “invite” button. If you invite a friend and they actually download the app and login, you get 1 billion tokens. I invited a few friends to test this. Sure enough they gave me 2 billion context tokens instantly. For heavy users, the real question is how much usage is that exactly? Short answer, I don't know. But I told max he needs to find a way to start making money before that runs out. As a moderate to heavy user, I'm assuming that will last till November (2026).
Click here for access to my invite code to start with 1 billion context tokens instead of 500,000,000
So the big question, is Meta's muse agent worth it? Yes. But how useful it will be is ultimately up for debate.
P.S. muse agents have an avitar at the top of the iPhone app or web browser. It starts out as a generic/friendly looking robot, but you can easily change it. So max looks like a disheveled nerd now. That's it for today, cheers.