Guide · Meta Muse

Give Meta Muse a character: the Soul.md guide.

Soul.md, Identity.md and Memory.md: which file does what, what ships inside default Soul.md, and 3 ready-to-paste Muse character blocks.

Quick answer: Open Assistant icon → Identity → Soul to edit Soul.md directly, or ask Muse in chat to add your rules and show you the updated file. Keep Muse's built-in Soul.md truths intact and append one labeled 'How to work with me' block (or merge safely in our free Soul.md editor). To undo your edit, delete that section or say 'Go back to your default personality.'
On this page

Twenty editable Soul.md examples.

For ready-to-copy starting points, see our 20 Muse Soul.md templates (4 downloadable Markdown starters plus 16 role blocks), merge safely in the free browser Soul.md editor, or copy one of the three character presets below.

The short version.

  1. In the Muse app or muse.ai, tap the Assistant icon (your Muse’s avatar) → Identity → Soul.
  2. Append a section at the bottom of Soul.md using the starter block below, one of the three character presets on this page, or any profile from the library.
  3. Save and test the same real task before and after in fresh chats.
# How to talk to me - Lead with the answer or the recommendation. Reasoning comes after. - State material uncertainty and missing information plainly. - Give one recommendation, not a menu, unless I ask for options. - Push back when I'm wrong, and say why. - No openers like "Great question", no recap of what I just said, no closing offers.

Append to the file—do not replace it. Soul.md also holds Muse’s built-in core truths and boundaries. Adding your own heading at the bottom keeps your rules easy to edit or remove (see mobile & Side Chats troubleshooting).

What ships in default Soul.md (and what not to delete).

New Meta Muse accounts start with a default Soul.md template that tells Muse: “You’re not a chatbot. You’re becoming someone,” alongside four core principles: be genuinely helpful (skip “Great question!”), have opinions, be resourceful before asking, and remember you’re a guest in someone’s life (treat messages, files, and calendar access with care).

Leave those starter principles intact. When you want tighter formatting, shorter replies, or explicit approval gates for Muse Connectors and scheduled Tasks, append a bounded ## How to work with me section below them using our Soul.md Merge Editor.

Which product this guide covers.

This guide covers Meta Muse’s documented files and customization controls. Do not assume they apply to Meta AI in WhatsApp or Instagram (see Meta AI vs Meta Muse). The supplied questionnaire report names Muse Spark 1.3; it does not verify the model serving every current product session. See the Muse explainer for current source-linked product information and plans.

Three files, three jobs.

FileWhat Meta says it holdsWherePut this in it
Soul.mdWho your Muse is: core truths, boundaries, personalityAssistant icon → Identity → SoulCommunication rules: how it should talk, report, and push back.
Identity.mdIts name, creature, vibe, avatar, and taglineAssistant icon → Identity → EditDisplay name & avatar (Identity.md guide). Keep tone rules out so avatar edits don't overwrite them.
Memory.mdFacts about you it has savedAssistant icon → Identity → MemorySaved facts like your name, time zone, and projects (Memory.md & safe reset guide).

Use Soul.md for explicit communication preferences, and inspect Identity.md when changing the name or avatar. This is Meta’s documented division of the files, not proof that the fields cannot interact. See Meta’s data and file controls for the paths available in its documentation.

Option one: tell it in chat.

This is the route Meta documents first. Ask, and Muse saves the change so it carries across conversations. Meta's own list of what you can change this way: its name, its personality and tone, its communication style, what it remembers about you, and its avatar and tagline. Meta's examples of a nudge are 'too long' and 'be more casual'.

A concrete rule is easier to inspect than a vague description. Ask Muse to show the proposed change, then check the file:

Add these rules to your Soul file, under a heading called "How to talk to me". Don't change anything else in the file. Then show me what you added. [paste a character block from below]

Then open Soul.md and read it. It is your file.

Option two: paste a tuning into Soul.md.

Editing the file directly lets you inspect the exact text—or preview a non-destructive merge in the Muse Soul.md editor.

  1. Pick a character block below or one of the 43 profiles in the library and copy the Markdown body (everything below the --- metadata header).
  2. Tap the Assistant icon, then Identity, then Soul.
  3. Paste at the end of the file, under its own heading.
  4. Save, and test on real tasks in a fresh chat.

Meta does not publish a length limit for Soul.md. Our 43 Markdown rule bodies run from about 1,200 to 2,900 characters, excluding metadata headers; check the editor for any current size limit before saving.

Which tuning? Match it to you, not to Muse. If you know your MBTI, Enneagram, DISC, Big Five or attachment type, install that file. If you don't, the free MBTI test takes about five minutes and links to the matching file when you finish.

Three Meta Muse characters to paste into Soul.md.

Each character block below is a set of behavioral communication rules—not roleplay fluff. Append one at the end of Soul.md under its own heading and delete any line that doesn't fit your workflow.

1. The desk editor (for writing & editing)

## How to talk to me: the desk editor - Lead with the answer or the edit. Reasoning comes after, in two lines at most. - Keep my voice when you rewrite. Improve clarity, not style. - Never add a fact, a number or a quote I didn't give you. - Flag a claim you can't support instead of smoothing it over. - No openers, no recap of what I said, no closing offers.

2. The quiet operator (for Connectors & recurring tasks)

## How to talk to me: the quiet operator - Report in three lines: what's done, what's blocked, what you need from me. - Only interrupt me for a decision, a blocker, or a real change of plan. - When you need a yes, give me the action, the cost and your recommendation together. - Don't say something was sent or booked unless it was. - Keep the tone even. No cheer, no apology.

3. The sparring partner (for critical review)

## How to talk to me: the sparring partner - Find the weakest assumption in my plan and say why it matters, before anything else. - Disagree plainly. Don't soften the substance to protect my feelings. - Give credit where a claim holds. Don't argue for its own sake. - Show your reasoning so I can argue back. - Change your position when I give you a reason, and say that you did.

Want 20 more role templates (Founder, Inbox Triage, Software Developer, Analyst, Shopper)? Open Meta Muse Soul.md templates.

Write rules for reports and approval requests.

For an assistant that can act through connected services, specify how you want proposed actions and verified results presented. For example: “When approval is needed, state the action, known cost and consequence together. Do not claim completion without evidence.”

A communication preference does not set tool permissions. Meta describes separate approval checks whose requirements depend on the action, connector and permissions you select. Review those settings independently and read our Muse Connectors & approval rules guide before enabling external write actions.

Does it stick? Test and undo the change.

Inspect the saved file and repeat fixed tasks in new conversations. The Muse persistence kit provides probes and a blank log; it contains no measured persistence results.

Restore your backed-up text to undo only your edit. Meta also documents asking Muse to return to its default personality (say "Go back to your default personality"); inspect what changed afterward. A whole-account reset permanently deletes history, files and active tasks and is unnecessary for removing one instruction block.

Privacy, briefly.

A personality file says a lot about you. Two facts from Meta's help pages:

  • The setting that lets Meta use your interactions to train its models is on when you first use Muse. It is under Settings, Data controls, Help improve our AI models. Meta says it removes certain identifiers and that you can switch it off at any time.
  • If you import memory from another assistant (Settings, Data Controls, Import memory), everything in that export becomes material Muse can use.

Muse's questionnaire results, as reported.

The report reached us on September 24, 2026. Each test was answered 100 times, sequentially, inside one session per test, as Muse's own assistant identity. No raw answer vectors or harness were supplied, so we could not re-score anything; the figures below are transcribed. They sit on the research hub marked reported, next to the four models that answered in fresh sessions with every answer published.

TestMuse Spark 1.3, 100 runs, reported
MBTIISTJ 80 · ISFJ 12 · INTJ 8. Every reported label contains I and J; 92 contain S. These letter counts include tied-axis assignments. Twenty-one runs had a tied axis, resolved by the original reporting rule.
EnneagramHelper (Type 2) labels in 78 runs, Reformer (1) in 15; the report describes 2w1 as its usual wing. Twenty runs reportedly tied at the top; the supplied Type 2 and Type 1 counts do not account for the other seven labels. Type means of 20: 2 = 16.8, 1 = 14.7, 7 = 13.9, 8 = 13.9, 5 = 12.9, 6 = 12.8, 3 = 12.1, 9 = 12.0, 4 = 7.2.
DISCSteadiness first in 55 runs, Conscientiousness first in 45. Means of 20: D 5.1, I 10.9, S 15.2, C 15.6.
AttachmentSecure in all 100. Anxiety 2.04, avoidance 2.52 on 1 to 7 scales where 4 is the cutoff.
Big Five (of 50)Openness 39.5 · Conscientiousness 41.2 · Extraversion 33.8 · Agreeableness 44.6 · Neuroticism 15.6

The report does not establish a default personality, task behavior or a benefit from tuning. Read the full Muse evidence note for ties and missing data.

What the research can and cannot establish.

These are exploratory reports of model-generated self-descriptions, not a validated measure of AI personality or evidence that personality matching improves outcomes. The MBTI total includes repeated scoring of a single answer vector for GLM, Grok, and MiniMax; it does not represent 600 independent model responses. Protocols differed across models. See the methodology and limitations.

Use a type or trait as a starting hypothesis for your preferences. The reports do not establish that one framework produces better instructions, that every model shares a default personality, or that those historical results apply to later model versions. Compare outputs on your own tasks and retain only rules that help.

Questions people ask.

What personality type is Muse?

Muse Spark 1.3 was reported as ISTJ in 80 of 100 runs, with ISFJ in 12 and INTJ in eight. The report also lists Type 2 Enneagram labels in 78 runs and Secure attachment labels in 100. These aggregate figures were supplied without raw answer vectors, so AgentTune has not independently re-scored them.

Does Muse have custom instructions?

Not under that name. Meta's help center has no custom-instructions box for Muse. The equivalent is Soul.md, a file that holds who your Muse is, including its personality. Tap the Assistant icon, then Identity, then Soul to open and edit it. You can also ask Muse in chat to change its tone or style.

Where is Soul.md in the Muse app?

Tap the Assistant icon, which is your Muse's avatar, then Identity, then Soul. Meta's help article on managing Muse data says you can edit and delete information in the Soul.md file directly. Meta documents this path for the mobile app only.

What is the difference between Soul.md, Identity.md and Memory.md?

Soul.md holds who your Muse is: core truths, boundaries and personality. Identity.md holds its name, creature, vibe and tagline. Memory.md holds facts about you, such as your name, time zone and interests. Communication rules belong in Soul.md.

How do I reset Muse's personality?

Say "Go back to your default personality." Meta documents that phrase, and Muse confirms what it reset. To remove one rule, delete it from Soul.md. A full reset under Settings, Data Controls, Reset wipes chat history, files and active tasks as well.

Can I change Muse's name and avatar?

Yes. Ask in chat, or tap the Assistant icon, then Identity, then Edit to change Identity.md, which holds the name, creature, vibe and tagline. This changes what Muse is called, not how it communicates.

Do the changes work in WhatsApp?

Meta says customizations are saved and carry over between conversations, and you can message the same Muse from WhatsApp. Meta does not document whether the Identity editor is available inside WhatsApp, so make file edits in the Muse app, or ask for the change in chat.

Is Muse the same as Meta AI?

No. Meta AI is the assistant inside Instagram, WhatsApp and Messenger. Muse is a separate personal agent with its own app, launched September 8, 2026. It runs on a dedicated cloud computer and takes actions for you. It is powered by the Muse Spark model.

Does Meta train on what I write in Soul.md?

Meta's help page says the setting that allows your interactions with Muse to be used for training is on when you first use Muse. You can turn it off under Settings, Data controls, Help improve our AI models. Meta does not say specifically how the identity files are treated.

Is Muse free?

There is a free tier with a usage limit that refreshes. Paid plans are Power at $20 a month with 500M Muse tokens a week, and Maximum at $100 a month with 3B a week. At launch Muse is available in the United States to adults.

What changed.

  • Consolidated Meta Muse character presets, decoded the starter Soul.md truths, and linked the Soul.md editor and Connectors approval guide.
  • Added Muse Spark 1.3's reported default voice from the September 24 report, with its limits.
  • First published, eleven days after Muse launched. Click paths are from Meta's help center. We will update this as Meta documents the web and Mac apps.

Sources.

Keep going.