2026.11.10
Why I went from dots and Grok Bot to using only Muse: after comparing them on four axes, Muse was the one chosen
With Muse, I can hand off the work I used to give to dots and Grok Bot to just one agent, quickly, and in large amounts. Comparing usage allowance, accuracy, response speed, and whether it does what I intend, Muse is good on all of them. That said, if you mainly use X, sticking with Grok Bot is fine.
Introduction
I usually had dots and Grok Bot manage my daily tasks and schedule. Since I recently started using Meta’s Muse, I’ve been using only Muse, so I’ve put together the reasons and background.
What I’ve felt from using it is that, with Muse, I can hand off the work I used to give to dots and Grok Bot to just one agent, quickly, and in large amounts.
If you’re using dots and feel it’s “slow” or “doesn’t work the way I want,” if you feel Grok Bot’s usage allowance isn’t enough, or if you’re curious about Muse, please take a look.
Conclusion: after comparing them on four axes, Muse was the one chosen
When I choose a personal agent, I look at four things: how much I can use it (usage allowance), whether its replies are accurate, whether its replies are fast, and whether it can do exactly what I have in mind.
Comparing them on these four, here’s how it turned out.
Grok Bot: its usage allowance wasn’t enough, so I took it off my list of options on that axis alone
dots: its replies are a bit slow, and even more than that, because it puts so much weight on safety, it often doesn’t work the way I want, so I stopped using it
Muse: I feel it’s good on all four
That said, if you mainly use X, sticking with Grok Bot is fine. If you use Instagram, I think Muse would be good too!
What is Muse?
Muse is Meta’s personal AI agent, and it became available in Japan in October 2026. In the chat in the app or on the web (muse.ai), you can hand off work just by talking to it as if you were sending a message to a person. Each user has a dedicated computer in the cloud, and Muse gets work done there by operating a browser, writing code, and so on. Think of it as Meta’s version of dots and Grok Bot.
Source: Meta, “Meta begins offering personal AI agent ‘Muse’ in Japan” (translated from the Japanese title)
Why I use Muse
From here, comparing it with dots and Grok Bot, I’ll introduce seven reasons why I’ve ended up using only Muse.
You can try it for free, and paying for it costs less than Grok Bot
You can start using Muse for free. Many of the basic features are free, and there are paid plans for people who want to use it more.
On November 7, a new model derived from the paid plan’s Muse Wonder rolled out to all users. The probability of Muse forgetting important information has dropped to 1/2.4, and its ability to see daily tasks through while correcting its own course has also gone up 67%. You can use the improved model even on the free plan.
As for paid plan prices, Muse also lets you start at a lower monthly price. If you sign up with monthly billing on the web, Muse’s Power is $20 a month and Maximum is $100 a month. The plans that include Grok Bot are Premium at $30 a month, Super at $100 a month, and Ultra at $200 a month. Muse’s official pricing page, Grok Bot’s official pricing page
Comparing the cheapest paid plans, Muse is $10 a month cheaper than Grok Bot, so you can start at about two-thirds of the price. And after trying it for free, when you want to hand off more of your daily tasks, you can increase your usage allowance starting at $20 a month.
Note: prices are the officially listed prices as of November 9, 2026, shown in US dollars. Grok Bot’s prices exclude tax.
On a paid plan, it’s accurate and replies fast
While the replies from dots are accurate, I sometimes had to wait a bit for them to come back. On Muse’s paid plan, for the way I use it, replies were relatively accurate and fast, and the tasks I asked for moved along at a good pace.
I can also check the progress of tasks running in the background, so I felt comfortable handing off tasks.
It has a full set of connectors, and you can put your own apps in Spaces
Muse has a full set of connectors for the services I use regularly, such as Linear, Notion, Gmail, and Google Calendar.
On top of that, Muse has Spaces. Like the dashboard in Grok Bot and Spaces in dots, it’s a feature that lets you keep apps you’ve made or simple web apps inside Muse in a form you can use right there.
If you tell Muse what you want to make, it builds it for you, and you can also fix it while using it. Information from services connected via connectors is also updated automatically inside the Space.
Source: Meta, “Design the Muse App, Just for You”
As an aside, here’s what my list of Spaces looks like.
The screen below shows the Space for my itinerary app “TabiClip” opened in the Muse app.
You can use it to some extent for free, and if you pay, you have plenty of usage allowance
You can use Muse to some extent for free, and if you pay, you can use quite a lot. The Power plan I use gives 500 million tokens a week. My usage allowance was reset when I paid, so if you’ve used up the free tier, you can pay and immediately ask it to continue.
It’s not nearly unlimited like dots, but in terms of being able to use a lot, Muse is strong too.
It makes its own decisions and acts, and doesn’t miss anything even in long texts or rough requests
Muse is highly capable of making its own decisions and acting on them. Even if I throw a long text at it as is, it doesn’t miss anything, and even when I ask roughly, it often fills in the context and acts. dots is smart too, but I feel Muse is better at making its own decisions and moving things forward.
With dots, there were times when it failed to keep its context in order and made decisions by confusing a previous conversation with the task in front of it, but with Muse that doesn’t happen, which I find appealing.
Also, in the Muse app installed on my Mac, under “Computer use” in Settings, I’ve set “Computer control,” which lets it operate the Mac, to “Always allow.” It doesn’t stop for confirmation every time, so I can hand off work on the Mac as is too. It’s reassuring to decide how much to hand off based on how you use your Mac.
Connect to your home Mac with Tailscale, and you can call Claude Code, Codex CLI, and Grok CLI from Muse
With Tailscale (a VPN service that securely connects devices to each other), Muse can SSH into your Mac at home. The key point is that it can SSH in without opening any ports and use the CLIs on that machine directly from Muse. I connect to my Mac Studio and call the following three from Muse.
Claude Code (Claude CLI): I hand off implementation to it or use it as a sounding board
Codex CLI: I create Codex threads and hand off implementation
Grok CLI: it searches posts on X and fetches the latest information
By calling Grok CLI, Muse can also handle information from X, which was Grok Bot’s strength.
Also, when I connected Tailscale, I had an experience where it didn’t connect properly because there was no “Tailnet Lock signature.”
If that feature is enabled, Muse’s device shows as “Connected” and also appears in the device list, but all traffic from unsigned devices is actually blocked, and the admin console shows “Locked out” for that device.
When it won’t connect, I think it’s a good idea to check the Tailnet Lock signature first, before suspecting permissions or the firewall!
Moving from dots or Grok Bot can be done just by having it load a GitHub repository
I’d had dots and Grok Bot push information about sessions, messages, context, and Automations to a private GitHub repository every night. Thanks to that, I was able to carry over all the background so far just by having Muse load this repository.
The only real hassle in switching was re-registering connectors and credentials. If you’re thinking about switching, I recommend having the AI assistant you use now regularly write out your settings and conversations to a repository.
Muse use cases
Here are the things I actually hand off to Muse.
Making a travel itinerary booklet from Instagram posts
Since Muse is a Meta service, you can connect it to Instagram and use information from posts you’ve saved.
You can even have it put together a Nara itinerary as booklet-style HTML based on places you want to go that you found on Instagram!
What I did with dots and Grok Bot can also be done with Muse
I usually speak ideas that come to mind and have it register them in Linear, and have it organize my ToDos every morning with an Automation. I also use it for things like adding event announcements to Google Calendar, getting notified of important emails in Gmail, and having it create presentation materials for talks. In my previous article, I also introduced a routine for receiving the latest information from X, turning articles into radio-style audio, and calling Codex and Claude Code via Tailscale.
For details, please check out my previous article. Except for the X routine, you can basically do the same things with Muse.
A list of things I’ve actually tried handing off to Muse
Here’s what I’ve actually tried handing off to Muse.
Finding the entry pass for an event I’m attending that day in Gmail
Taking the measurements I made of furniture and appliances, such as a bed and a refrigerator, as is, and transcribing them onto a Notion page
Here’s what it looked like when I had it transcribe the sizes.
Having it update, on its own, the background information it should remember (memory) based on casual conversation
Building and publishing “VINYL SHELF,” a collection app where you browse a record shelf, pick a record, and listen
Here’s the public link to “VINYL SHELF.”
https://muse.ai/s/rekodopeng-apuri-xcn6xkpxexz6xoxb6/izFFgvKA4bfx60KkARuwzAsJQ8Rl7klIz28am0kgJf8
Creating a selfie-style image and turning it straight into a video
On a paid plan, image and video generation is about the same quality as Grok too.
Working with it to pin down why Tailscale wouldn’t connect (Tailnet Lock), and connecting to my Mac Studio over SSH
Calling Grok CLI on my Mac to gather my X posts to use in this article
Bouncing ideas off Claude Code on my Mac to organize the structure of this article
Calling Claude Code from Muse and organizing my thoughts while bouncing ideas off it
Especially handy is bouncing ideas off Claude Code. When I ask Muse to bounce ideas around, it goes back and forth with Claude Code (Claude CLI) on my Mac several times to refine the ideas, and then organizes them for me. For this article, Muse showed its proposed structure to Claude Code and got suggestions for missing items, the order, and headings. I think this is a way of using it that’s only possible because Muse can call the CLIs on my Mac.
Things to watch out for when using Muse, and how to choose between it and dots or Grok Bot
For long texts and difficult tasks, dots is good
dots is very smart and its usage allowance is nearly unlimited, so you can hand off quite long texts and fairly difficult tasks.
On the other hand, as mentioned earlier, its context can get mixed up, making it hard to get the output I intended. Requests that contradict rules decided earlier can get refused, and I can’t fix the context from my side either. Because it prioritizes safety, its actions tend to get blocked, and there are cases where it won’t act even when I give permission, which also makes it a little hard to use…
This difficulty of use is the main reason I moved from dots to Muse.
If you mainly use X, sticking with Grok Bot is fine
I think Grok Bot’s only weakness is its usage allowance. Including the ability to use information from X, its ecosystem and UX are overwhelmingly strong, and I feel they’re better than Muse’s.
However, since the usage allowance was reduced, I can’t use it all that much even when paying quite a lot, and sometimes I use it all up in a single day. Because of that, I started hesitating over whether “I should just do this task myself.” The original appeal of a personal agent, being able to casually hand off anything, was fading, so I stopped using it.
Hand off the routine for receiving the latest information from X to Grok CLI
A routine for regularly receiving the latest information from X can’t be reproduced as is in Muse. If you’re not using Grok Bot, I recommend handing off the work of fetching information from X to Grok CLI. I also fetch it by calling Grok CLI from Muse.
Check first whether Muse has connectors for the services you want to use
dots and Grok Bot have more connectors. If there’s a service you want to use, check first whether it’s among Muse’s connectors.
If you’re going to pay, subscribe from the web, not the app
The Power plan is 3,499 yen a month through the app. From the web (muse.ai), it’s 2,799 yen a month, 700 yen cheaper, and I subscribed from the web too.
On a Mac, turn on “Run at startup” to keep Muse running
In the Muse app installed on your Mac, open Settings, select “General” in the list on the left, and turn on “Run at startup” under “App behavior.” Once it’s on, Muse opens automatically when you log in to your Mac. If you use it as a personal agent, I recommend keeping it running. I have it turned on too.
You can turn off learning in Muse’s settings
“Help improve AI models” under “Data controls” in Settings is the option that allows your interactions with Muse to be used to develop and improve Meta’s AI. If you’re concerned, you can turn it off.
Closing thoughts
Comparing usage allowance, accuracy, response speed, and whether it does what I intend, Muse is good on all of them
You can try it for free, and even when you pay, it’s cheaper than Grok Bot
If you connect to your home Mac with Tailscale, you can also call Claude Code, Codex CLI, and Grok CLI
If you mainly use X, sticking with Grok Bot is fine
If you’re struggling with “I’m using dots, but it’s slow and doesn’t work the way I want,” I recommend Muse. If you feel Grok Bot’s usage allowance is too small, I think it’s worth giving it a try too.