Personal Intelligence System for creative professionals. Open-source, MIT-licensed, runs locally on Obsidian + Claude Code. Vault scaffolding + 14 AI skills + 6 ingestion tools that turn your scattered archive into a queryable second brain.
Not a notes app. A layered system that archives source material, builds structured knowledge, tracks projects, and generates outputs — with AI-powered skills that do the heavy synthesis.
"Most AI tools wrap a chatbot around a prompt. This wraps a chatbot around your archive. Output sounds like you because it is you."
git clone https://github.com/mattymostudio/creator-system.git
cd creator-systemOr Code → Download ZIP above, then unzip wherever you keep projects (~/Documents/, ~/code/, anywhere). The folder you end up with is your working vault — no install step, no second copy.
- Obsidian — free, the editor for your vault
- Claude Code — the AI layer; free tier covers the first session, Claude Max (~$100/mo) recommended once you're loaded up
Claude reads CLAUDE.md, sees you're a fresh user, and walks you through:
- Personalizing the vault with your name + practice
- Dropping your first source into
vault/01_INBOX/To Process/ - Running
/ingestto process it - Running
/diarizeto build your first canon page - Initializing git so everything you do is undoable
Total: ~10 minutes to a working personal intelligence system you own.
Tip: open this same folder in Obsidian alongside Claude Code. You'll see the markdown files Claude is writing in real time, and you can edit any of them directly.
creator-system/ ← you're here; this IS your vault
├── vault/ ← 10-folder layered knowledge system
│ ├── .claude/commands/ ← 14 AI skills
│ ├── 00_HOME/ ← Navigation hubs and system state
│ ├── 01_INBOX/ ← Drop new material here
│ ├── 02_SOURCES/ ← Raw source material (immutable)
│ ├── 03_SOURCE_NOTES/ ← Processed extracts
│ ├── 04_CANON/ ← Synthesized knowledge (the wiki)
│ ├── 05_PROJECTS/ ← Project tracking
│ ├── 06_OUTPUTS/ ← Reusable deliverables
│ ├── 08_RESEARCH/ ← External topic knowledge
│ ├── 09_IDEAS/ ← Ideas pipeline
│ └── 10_META/ ← System rules + templates
│
├── tools/ ← 6 ingestion pipelines
│ ├── photo-processor/ ← iPhone photos → events, faces, locations
│ ├── takeout-processor/ ← Google Takeout → contacts, timeline
│ ├── chatgpt-ingest/ ← ChatGPT export → topic index
│ ├── facebook-ingest/ ← Facebook archive → message threads
│ ├── press-ingest/ ← Press URLs → cleaned local archive
│ └── granola-ingest/ ← Granola meetings → rolling source-note
│
├── README.md ← You are here
├── CLAUDE.md ← Instructions for Claude Code
├── RECIPES.md ← 23 step-by-step things to try
├── Standard Operating Procedure.md ← Daily use, weekly maintenance, manual fallbacks
└── Data Sources to Gather.md ← What raw materials to collect
Material flows through layers, gaining structure at each step:
raw source → source note → canonical knowledge → project / output
Not every source becomes a note. Not every note becomes canon. The system is selective by design — fewer, better pages.
Sources have trust levels. When they disagree:
- Journals win for emotional truth and lived experience
- Business documents win for dates, amounts, and what actually happened
- Press tells you what others think happened — useful, but not ground truth
- Your interviews tell you what you chose to say publicly — performance, not diary
| Command | What it does |
|---|---|
/ingest |
Process a raw source through the full vault pipeline |
/diarize |
Build a comprehensive page from all sources about a subject |
/enrich |
Fill out a thin page by finding all related sources |
/vault-lint |
Run 10 structural checks and write a report |
/improve |
Analyze patterns and propose system improvements |
/emerge |
Surface ideas the vault implies but never stated |
/challenge |
Pressure-test a belief against the vault's own evidence |
/connect |
Find hidden bridges between two topics |
/drift |
Compare stated priorities vs actual activity |
/ideas |
Vault-wide brainstorm grounded in what the vault knows |
/council |
12 professional personas evaluate a project or idea |
/recap |
Close out a session with a dated "what changed / next / loose ends" entry |
/autoresearch |
Autonomous research loop on an external topic |
/granola |
Synthesize Granola meeting pulls (pairs with the granola-ingest tool) |
You don't need to type the slash. Plain English works — "build me a canon page about Devon" runs the same path as /diarize "Devon".
RECIPES.md— 23 step-by-step recipes (input → tool → skill → output) by data type. Open this first.Data Sources to Gather.md— What to collect and where each type goesStandard Operating Procedure.md— Daily use, weekly maintenance, manual fallbacksvault/10_META/AGENTS.md— The vault constitution: page types, naming conventions, source hierarchy, frontmatter schemas
The system works entirely by hand. Open vault/00_HOME/Start Here.md in Obsidian and follow Standard Operating Procedure.md. Every workflow has a manual equivalent — AI just makes it faster.
MIT — see LICENSE. Attribution to Matty Mo Studio appreciated, not required.
Status: Early version, currently in private beta with Cohort 01 of the Creator System cohort program. Active development. The kit is yours to use — the system is opinionated, but the data is always yours, on your laptop, in plain text.