Effigy

A structured self-knowledge practice built on Claude Code and markdown files. Feed it your words — voice memos, conversations, writing, therapy transcripts — and it builds a model of who you are that exceeds your own self-knowledge.

Not a journal. Not a personality test. Not an AI therapist. A living portrait that gets sharper every time you use it, driven by questions the model generates about what it doesn't know yet.

Who This Is For

People already doing the work. You journal, or you're in therapy, or you think about yourself structurally, or you've hit the ceiling of what unstructured self-reflection can do. You don't need to be convinced to be vulnerable — you need a better instrument for the self-examination you're already doing.

How It Works

The system runs on a loop:

1. /setup — First time only. The system asks you foundational questions. You talk. It builds your initial model.
2. /ask — The model surfaces questions about you — things it wants to know, tensions it can't resolve, gaps it's identified. You answer them.
3. /ingest — Your answers (and any other material you drop in) get processed into structured notes.
4. /synthesize — Notes get folded into the model. New questions are generated. The portrait sharpens.
5. /ask — The questions are better now. Repeat.

You can also run /interview [name] to generate customized questions for someone who knows you. Their perspective is the corrective to your self-report — and the point where the model starts saying things you didn't know about yourself.

What You Get

• Soul Card (portrait/card.md) — One page. The densest possible capture of you. The thing you share.
• Model files (portrait/model/) — Disposition, drives, values, defenses, relationships, narrative, formulation, contradictions. The depth.
• Voice (portrait/voice/) — Raw quotes and linguistic fingerprints. The model should sound like you.
• Snapshots (portrait/snapshots/) — Dated portraits that accumulate. How you change over time.

Quick Start

git clone [repo-url]
cd effigy

Run /setup in Claude Code. Talk for 20 minutes. You'll have a model.

See docs/getting-started.md for details, and docs/how-it-works.md for the methodology.

Requirements

• Claude Code (CLI, VS Code extension, or desktop app)
• A way to talk instead of type. This system works best when you speak — typing changes how you think. Free options:
  • SuperWhisper (Mac) — speech-to-text that types as you talk
  • iPhone Voice Memos — records and auto-generates transcripts you can paste in
  • Any speech-to-text tool that gets your words into text

Privacy

Everything is local. Markdown files in a git repo you control. Nothing is sent anywhere you don't send it.