soul.md
Your identity, not theirs.
Every new AI chat starts from zero. soul.md fixes that. One plain file that tells any AI who you are, how you want it to talk, and what you're working on. You write it once. You own it. It works everywhere.
Where this is going
AI coding was wave one.
AI work management is wave two.
Wave three is humans building personal AI — portable career capital.
The thing you're soon unhireable without.
How it works
Three steps, no terminal required
01
Answer a short interview
Paste one prompt into the AI you already use. It asks twelve questions and writes your soul.md for you.
02
Give it to your AI
Paste the file into a ChatGPT or Claude project — anywhere your AI takes instructions. Developers can wire it in with one command.
03
Every chat knows you
Your tone, your rules, your current work. No more introducing yourself to a machine, again.
One file is the point: when your work changes, you update it once and every tool follows. Pasted, that's a quick re-paste. Wired in with the developer tools, it loads automatically — and the AI keeps the file fresh for you as you work.
The file
Plain text, yours forever
Something you can read, edit, and keep for as long as you like. No settings screens, no vendor account. It looks like this:
Every AI app keeps its own memory of you, in its own cloud. soul.md is the one file about you, in your hands — and it moves with you when you switch tools.
> Format: soul.md v2
## /identity
Maria. Head of brand at a two-person studio.
## /voice
Direct, no fluff. Answer first. Push back when I'm wrong.
## /now
Client rebrand due Friday. Learning motion design.
## /preferences
Short answers. No jargon. Options, then a recommendation.
The difference
Isn't this just ChatGPT's memory?
Built-in memory is genuinely good: zero effort, always on. If you use one AI and don't mind it holding the file on you, it's enough. Here's what it can't do:
Yours
Vendor memory lives in their cloud, attached to your account. Cancel, switch, lose access — gone. soul.md is a file on your computer. Read every word, edit it, delete it, put it in git.
Every AI, not one
Claude's memory works in Claude. ChatGPT's works in ChatGPT. Use three tools and three strangers keep three different files about you. soul.md is one identity, everywhere.
Stated, not inferred
Memory collects crumbs from what you happened to say, and injects them without showing you the whole picture. soul.md is deliberate: your voice, your rules, your current work. You approved every line.
Outlives the product
Models change and memory formats are opaque. Plain text survives every platform decision made by someone who isn't you.
Not either-or: keep built-in memory for recall. soul.md is the durable part of you — and you should own it.
For developers
Three commands, one config block
AGENTS.md describes your repo. soul.md describes you.
The CLI
npx manoma init # create your soul.md npx manoma interview # an AI you already use writes it with you npx manoma compile # wire it into Claude Code npx manoma refresh # your AI proposes updates — you approve npx manoma remember # seed the AI's own memory from your file
compile adds a managed block to Claude Code's config with a live import, so every new session starts already knowing you.
The MCP server
Add manoma-mcp to Claude Desktop, Cursor, or any MCP client and the AI can write to your file as you work. Say “note this decision” and it lands in your soul.md, dated — every change a git diff you can read.
When your file goes stale, the AI notices and offers an update — nothing is ever written without your approval, and every AI edit is dated and signed.
One config block. Copy it from the docs →
Open format, public domain, one-page spec. Docs · Spec · GitHub
Early days. If something feels off or confusing, tell me. Feedback is the whole point right now.