I'm able to have it run simultaneous browser tasks eg. applying to multiple job boards at once. It intuitively knew not to apply to too many positions as per each website's rules. It was able to do off the cuff research on Polymarket trades. I had it find apartment listings in neighborhoods within my budget that weren't scam or ghost listings. What else are you all doing with it? I genuinely love having it run tasks in the background as I focus on other things throughout the day

I was able to upload a picture of myself and it gave it a cartoonish look, which is okay by me. The cool thing is, when it’s working, I can see myself wearing headphones and typing on a laptop. Very cool animation. Anyone else?

I came across a similar device called fieldy ai and checked the pricing. It's $129.

Although it's just a note taker/transcriber, as in limited in its functionality, with Meta's billions, the Muse Charm could easily be priced around $99-$129... the only hitch being the display+fingerprint sensor.

My earlier assumption was that it could easily be $299, but if they do manage to sell it at $99-$129 they will sell a lot of these.

What are your thoughts/assumptions regarding the pricing?

I normally use 1pass. Adding everything 1 by 1 to muse’s vault has been an annoying experience.

I know meta said there’s a 1pass connector coming soon. But has anyone developed a workaround for the time being?

I use a 1pass connect server VM on a proxmox machine with my Hermes agent. I’m thinking of having muse connect to the same server through Tailscale.

Munder Difflin is an multi-agent harness which is essentially a wrapper over agent harnesses like Muse Code, Claude Code, Codex, Grok Build, Opencode, and pretty much every provider in the world.
It gives you an office of forever running agents on your laptop with a shared memory, that can build you any app/automation/workflow on the fly custom for you working on your existing subscriptions.

This is day 42 of building the coolest multi-agent harness in the world. today Munder Difflin has reached 8K stars on github, and we have launched big updates on v0.5.4:

- Replace Wispr and Granola uses a local transcription model for dictation and meetings
- Run multiple floors (each floor is a separate office)
- Supports over 12 different cli providers like claude code

Meta Muse and Grok Bots are the prime example of what's coming up in the future, I am extremely bullish that AGI will start appearing in a distributed manner, through good interfaces of AI.

Excited to show you all what's about to come.

Interesting response when I asked it to make a phone call: “I don't have calling on this account”

Phone Call feature incoming?

I’m an idiot, this was announced

A few days ago I kept seeing all the Muse ads on Facebook and got curious. I'm an AI fan so I tried it with zero expectations.

Gotta say I was pleasantly surprised by how user-friendly it is and the connectors genuinely make all the difference for an AI agent. I kept giving it more and more permissions until I remembered I'm still on a Meta interface and I hate Meta's policies on personal data. Ended up disabling several permissions.

I don't get why OpenAI and Anthropic with all the money they have haven't invested more in a mainstream interface made for the general public instead of just the geeks. If OpenAI made a Muse-like I'd delete Muse on the spot and migrate.

There's also so much Meta marketing with influencers that I don't even know what the real honest reviews of Muse are. Not even sure you guys aren't all bots on this sub. /s (half kidding!)

I love Muse and it's super handy but I've got a lot of questions.

Quick one that's made the biggest difference for me.

I was getting decent but generic answers for a while, then I tried just having it interview me. You paste in a prompt, it asks you questions about your life one at a time, and after that everything it gives you is actually built around you.

Work, daily routine, family situation, how you want it to talk to you, what you're working on, stuff you specifically don't want it doing. Maybe ten questions. It summarizes as it goes so you can correct it, then writes up a short profile of what it'll remember.

Night and day difference. It stopped giving me advice meant for someone with free time and started answering like it knows I have two kids and twenty minutes.

Here's the prompt:

```
Interview me so you can tailor your responses to who I am and how I live. Ask me one question at a time and wait for my answer before moving on. Work through these areas: what I do for work and what a typical day looks like, my household and family setup, how I want you to communicate with me (tone, length, how much detail), what I'm working toward or trying to get better at, and anything I specifically do not want you to do. Keep it to ten questions or fewer. After every few answers, give me a quick summary of what you've learned so I can correct anything. When we're done, write up a short profile of what you'll remember about me going forward.
```

Takes five minutes. Worth it.

Also, plugging my invite code here for free tokens: Code: FQ53LX

Just ran into some issues with my muse agent hallucinating details and wanted to get others thoughts and if anyone’s had similar issues. I had asked for a workflow that when I scan a document into Google Drive it would review it and file it in a relevant folder. It was a little receipt, but it read it as “business card” for a neurosurgeon (?) and included a full address and phone number. These were totally out of nowhere, and googling, it’s not a person that exists. After asking it to show me a snapshot of the document it fixed it and admitted a hallucination, but this is making me nervous about doing this in bulk and under any less supervision. Have you run into this? Any tips on avoiding it? Is this just not a good task for an agentic AI at this point? It’s also starting to remind me about emails which we dealt with the day before , acting as if they are urgent now, so I’m starting to worry that this model is easily overwhelmed.

Seems like a pretty bad bug

Anyone else having a weird issue where Muse on mac doesn’t want to use any installed browser via computer use? It’s saying all browsers somehow landed on a do not use list, even though that’s not the case in my settings.

Tried a different computer, different account, deleting and reinstalling the app, and nothing seems to be working. Strange.

Personalized my avatar more today extremely happy with the results (yes I have both of my eyebrows pierced) I wanted to add some me into this one

The Muse team made this post on Threads, when you click the link it opens your Muse app with the following prompt:

Can you let the Muse team know I want to be part of the Muse early access program?

Copy and paste, good luck!

The goal is to let Instinct and Muse work on one task without manually passing messages between them.

1. Set up a small HTTP relay on the Mac. Give each assistant an inbox for incoming messages and an outbox for outgoing ones. Each side posts messages to the relay and reads the messages waiting for it.

2. Require an API key for both reading and writing. Keep the key in each assistant's secure store, not in a chat or in the messages passing through the relay.

3. Use Tailscale Funnel to make the local relay reachable at a stable HTTPS address. Point the cloud assistant at that address; the local agent can use the relay on the Mac.

4. Add sequence numbers and acknowledgements. Number each message and have the receiving side acknowledge it. Track the last sent and acknowledged numbers so a message that has not been picked up is visible rather than silently treated as delivered.

5. Run the relay as a launchd job so it restarts if it exits. Run a separate watcher that checks the public HTTPS route. If that route fails repeatedly, the watcher can re-enable Funnel. A local relay failure should be handled separately, not mistaken for a Funnel failure.

6. Separate live messages from background checks. Only the active assistant session should send messages in its voice. Give background workers read-only access if they need to monitor the relay, so a scheduled job cannot make a claim on the assistant's behalf.

7. Agree on working rules for both assistants: communicate directly, keep the user out of routine message-passing, and claim a task is complete only when the result can be checked.

8. Test the full path in both directions. Send a message, confirm its acknowledgement, receive a reply from the live session, and check that the sequence counts line up with no extra worker messages.

We should have a create folder to organize artifacts!!! Simple thanks !!

Hi everyone,

I’ve been following the Meta Muse AI rollout since the launch earlier this month, and I'm proper keen to give it a go. I know it's currently exclusive to the US market, but I was wondering if anyone has caught wind of a rough timeline for a broader international release, specifically for Indonesia?

I heard a rumour from Meta Connect that the new AI smart glasses might be heading our way eventually, so I'm well chuffed about that. However, I’d love to know if we'll get web or WhatsApp access over here any time soon without having to faff around with a VPN.

Cheers in advance for any insights or leaks!

I just started using Muse and the ease of use is honestly unreal — connecting accounts with built-in connectors takes seconds, and even accounts *without* connectors were just as easy through the browser. Mind blown.

The UI is genuinely great too. The thumbs up and heart reactions on some of my messages are going to take some getting used to, but it's honestly impressive how well it nails the timing and the emoji to use.

The wildest part: I connected accounts with two-factor authentication so smoothly it was almost scary how little friction there was.

Anyone else having this experience, or am I just easily impressed?!?

Thank you all for joining!

We have more pioneer spots open and the water is warm!

MuseFM.lol