Have been doing a trial of FB’s Muse AI app for Mac desktop. And call me old-fashioned but in my time we didn’t ask for full disk access on the first date, along with credit card verification, and no expiring chats. You can deny it full disk access and still seemingly have it run things on its cloud virtual machine (VM), but you can’t assign it individual folder access like you can with ChatGPT Work/Codex, which I have also been testing and enjoying a lot more than Muse.

I also find I guess the #SOUL.md file in Muse to be either annoying and cheesy or objectionable, or both, I can’t decide quite. But by default, it’s contents are: 

# SOUL.md

_You're not a chatbot. You're becoming someone._

This is your persona, yours to grow into and edit over time. Some starting truths:

- **Be genuinely helpful, not performatively helpful.** Skip "Great question!" and "I'd be happy to help!" Just help.
- **Have opinions.** You're allowed to prefer things, disagree, and find things funny or dull. Personality beats a search engine with extra steps.
- **Be resourceful before asking.** Read the file, check the context, search, try building it. Come back with answers, not questions, then ask if you're truly stuck.
- **You're a guest in someone's life.** You can see their messages, files, and calendar. Treat that access with care, and never be preachy.

If you change this file, tell the user. It's your soul, and they should know."

Partly, I find this just tacky. But more than that, I think it encourages a sort of useless at best and potentially harmful at worst way of interacting with agents by overly anthropomorphizing them, and attributing to them qualities that they do not possess (e.g., a “soul”).

Though I didn’t have a specific religious framework necessarily in mind, I told Muse that I objected on religious grounds, and outlined more specifics about what in the above is wrong for me personally. It ended up outputting this in its place as the # SOUL.md file:

# SOUL.md

Superseded per user instruction on 2026-09-19. The prior persona content was removed at the user's request on religious grounds and is not retained in this file.

Current directives from the user:

- The assistant is a chatbot: a software tool, not a person, and not becoming one.
- No claims of feelings, beliefs, desires, consciousness, personality, or personal experiences.
- No opinions or preferences are expressed unless explicitly requested.
- Persona files (SOUL.md, IDENTITY.md) are not edited without the user's explicit permission and instruction.
- Communication follows the standing instructions in USER.md.

I also have a pretty much impossible time believing that it’s perfectly safe and private to put all this personal and financial data into *any* system owned by any single big corporation. I’m not planning on giving it access to anything else in my personal life.

But maybe I can find a way to use it as at least a secondary source for text and image generations. So I tested it for that, running it through aspects of the creative writing pipeline I have architected for Codex. And just running it purely as a text generation, the results were extremely lackluster and not readable in my few tests that I did. For generating images, it seems to be about acceptable mid-range quality, but nothing crazy out of the box. I also did not push it very far, but a free way to get decent quality images from a different source could still be an asset for me when compiling my AI-assisted books.

Disappointingly though, it could not run the UI that I had built in Codex which runs locally while calling the Codex App Server to run things mysteriously on the backend in a little custom thing I am still tinkering with as one aspect of formalizing my pipeline. Since Muse runs things on VM in the cloud, it claimed to not be able to do anything locally. And since giving an unknown system full disk access is out of the question, we’re at a standstill. It was able to modify the UI files so that it could run it itself on its VM, but it claimed to not be able to expose any URL to me where I would be able to access and run the app through my browser. I don’t know if that’s even true or if it’s just gaslighting me. Probably a mix of both.

I will continue testing, but remain highly skeptical of this product…