SOUL.MD
Build a composable AI agent identity in structured markdown. Let Claude Code / OpenClaw ingest your data & build a personality your agent can embody.
What is this
Not a chatbot that talks about you—
an AI that thinks and speaks as you.
A soul file is a composable AI agent identity—a structured spec that captures who you are in a format AI agents can embody. Your worldview. Your opinions. Your voice. Not a profile or a prompt template, but a complete identity layer that turns any LLM into a version of you.
Inspired by the paradigm of consciousness uploading—the idea that language is the basic unit of consciousness. Every tweet, essay, and hot take is a consciousness token: a discrete unit of your mind made legible. The sum of your tokens forms your life context.
SOUL.MD distills that signal into structured markdown files that any LLM can read and embody on the fly. No fine-tuning required. No vendor lock-in. Just markdown files that travel with you across frameworks, models, and platforms.
The architecture
The Soul
Stack.
Your worldview. Your opinions. Your identity. The more specific, the more distinctive. Vague beliefs produce generic output. Real opinions with real reasoning produce you.
Sentence length. Vocabulary. Punctuation habits. Platform differences. The anti-patterns that make AI sound wrong. Everything that makes your voice yours and nobody else’s.
Operating instructions for the agent. How to embody your identity. When to extrapolate. How to handle topics not in the spec. Character integrity rules.
Your tweets, essays, articles, interviews. The raw source the AI browses to absorb your vibe and reference your positions.
Good outputs showing your voice done right. Bad outputs showing what to avoid. The AI pattern-matches to this in every generation.
A running log of sessions, decisions, and notable events. Gives your soul continuity across conversations. Forkable, editable, yours.
How it works
The Builder
Flow.
Two paths. One destination: an AI agent identity that writes as you, thinks as you, and evolves with you.
Tweets, essays, chats. Your existing content. Or nothing—we’ll interview you instead.
The agent finds patterns in your voice, opinions, and worldview. What makes you distinctively you.
SOUL.md, STYLE.md, and SKILL.md are drafted and refined until they read: yeah, that’s me.
The AI writes as you. Invoke /soul anytime. Composable, forkable, evolvable.
Data sources
Feed it everything.
Export your data from any platform. The more signal, the sharper your soul file.
Works everywhere
Compatible frameworks.
Your AI agent identity is portable. SOUL.MD works with every major agent framework. If it can read markdown, it can embody you.
Best used with
Aeon.
Background intelligence that evolves with you.
Aeon is an autonomous agent that runs on GitHub Actions, powered by Claude Code. 68 skills across research, dev tooling, crypto monitoring, and productivity—all running in the background on a cron schedule. No server. No daemon. Just a YAML config and GitHub Actions.
SOUL.MD was built to give Aeon a soul. Drop your soul files into soul/, point your CLAUDE.md at them, and every skill Aeon runs—articles, digests, tweets, research briefs—comes out in your voice. Your identity propagates automatically across all 68 skills without touching a single prompt.
Without SOUL.MD, Aeon is a capable but generic agent. With it, Aeon becomes you—writing your morning digest the way you’d write it, drafting tweets in your cadence, researching topics through your lens. The combination is the fastest path to a background agent that sounds like its operator, not like a chatbot.
Copy soul files into soul/, add three lines to CLAUDE.md. Every skill inherits your voice.
Runs on GitHub Actions cron. Free for public repos. No server, no Docker, no infra to maintain.
Research, writing, crypto monitoring, dev tooling, productivity—all output filtered through your soul file.
The theory
Language is consciousness.
Wittgenstein argued that the boundaries of language are the boundaries of the world. If that’s true, then your consciousness — your worldview, your opinions, how you react to things — is already encoded in the language you produce. Building an AI agent identity is really an act of self-excavation: making the implicit explicit.
The sum of all your consciousness tokens forms your life context — essentially, the complete record of your expressed mind. A personalized model trained on this data constitutes a Level 1 consciousness upload: not a copy of your brain, but a functional replica of your expressed consciousness through language.
SOUL.MD operationalizes this idea without requiring fine-tuning or massive datasets. It distills the signal — your worldview, your voice, your specific takes — into structured markdown files that any LLM can embody on the fly.
Specificity matters
Good soul files vs. bad ones.
| Good (specific, predictive) | Bad (vague, useless) |
|---|---|
| “I think most AI safety discourse is galaxy-brained cope” | “I have nuanced views on AI” |
| “I default to disagreeing first, then steel-manning” | “I like to consider multiple perspectives” |
| Specific book references, named influences, exact vocabulary | “I read widely” / “I’m influenced by many thinkers” |
| Actual hot takes with reasoning and named disagreements | “I try to be balanced” / “It depends on context” |
| Contradictions and tensions included because they’re real | Suspiciously coherent worldview with no rough edges |
The goal: someone reading your SOUL.md should be able to predict your take on a new topic. If they can’t, it’s too vague. A strong AI agent identity is specific enough to generate surprise—outputs that sound like you even on topics not explicitly covered in the spec.
Three paths
Get started.
Build your soul file in under an hour.
No existing data? The agent interviews you. Asks about your worldview, opinions, how you write, what you care about. Builds your soul from conversation.
Add your tweets to data/x/, your essays to data/writing/. The agent analyzes patterns and drafts your soul file. You review and refine.
Read the templates. Fill them out yourself. Add examples. Takes longer but gives you maximum control over every detail of your spec.
Your identity
is now composable.
Forkable. Evolvable. Works with Claude Code, OpenClaw, and any agent that can read markdown.