Get started in 2 minutes
Install, point at your notes, explore. Your half-tagged vault is enough.
1. Install
â–¶ Don't have a markdown vault yet?
Enzyme works on markdown files—readable in any editor, portable across tools, yours forever.
- → Obsidian — Already markdown. Point Enzyme at your vault.
- → Kindle/Readwise — Export highlights from Readwise
- → Voice memos — Transcribe with MacWhisper
- → PDFs — Convert with docling
- → Starting fresh — Create a folder, start capturing. Even 20-30 files is enough.
2. Initialize your vault
Navigate to your vault and run enzyme init.
Enzyme analyzes your vault structure (recent captures, recurring tags, folder patterns) and builds a graph of how your thinking connects.
$ enzyme init
What happens during init
- 1 Structure reading — Enzyme scans your tags, links, and folders—including incomplete ones. Your half-finished system is treated as intent, not failure.
- 2 Graph extraction — Entities and connections are mapped. Long notes become clusters; scattered captures reveal the threads between them.
- 3 Catalysts surface — Questions, themes, insights emerge from your own words—the threads worth pulling.
This sends excerpts to AI for analysis. Your raw files stay local. See privacy for details. For a deeper look at the Petri, Catalysts, and Entities, see the concepts documentation.
Providing your own guide
You can help Enzyme understand your vault by providing a guide — a short description of how you organize your thinking. This is especially useful for vaults with lots of structure, or if you want to steer the analysis toward certain areas.
$ cat VAULT_GUIDE.md | enzyme init
# Or pass it directly
$ enzyme init --guide VAULT_GUIDE.md
â–¶ What to include in a vault guide
Write it however feels natural. A few sentences about:
- Folder purposes — "inbox/ is quick captures, garden/ has evergreen notes"
- Key tags — "I use #to-review for things I want to revisit"
- Important people or concepts — wikilinks you reference often
- What matters most — areas you want Enzyme to know about
Your guide shapes which questions Enzyme surfaces from each entity in your vault. Mentioning a tension or a recurring theme tilts those questions toward what you're actually thinking about. See how the guide differs from an instruction file →
This isn't a template. Just jot down how you think about your notes. Enzyme uses it as soft guidance, not strict requirements.
â–¶ Using local or self-hosted models?
If you're running a local model server (LM Studio, Ollama, vLLM), set both the endpoint and model name before running init:
export OPENAI_BASE_URL="http://localhost:1234/v1"
export OPENAI_MODEL="qwen/qwen3-8b"
enzyme init
# Ollama
export OPENAI_BASE_URL="http://localhost:11434/v1"
export OPENAI_MODEL="qwen3:8b"
enzyme init
Both variables are required—the default model is an OpenRouter slug that won't work with local servers. OPENAI_API_KEY is optional for localhost.
For remote OpenAI-compatible providers (OpenAI, Together, Groq):
export OPENAI_API_KEY="sk-..."
export OPENAI_MODEL="gpt-4o"
enzyme init
3. Explore
Install the Enzyme plugin in Claude Code or Cowork. The plugin gives Claude a skill for exploring your vault — no MCP configuration needed.
$ claude plugin marketplace add jshph/enzyme
$ claude plugin install enzyme
Try asking:
- > hey what do you have going on in my vault today?
- > help me come up with 10 writing prompts based on my thoughts on travel and paradox
- > help me with wording in a product brief based on my Readwise annotations that describe my ideas about storytelling
Quick reference
| Command | What it does |
|---|---|
enzyme init | Analyze vault, build graph, find your threads |
Questions
What formats work? ▼
Markdown files (.md), Obsidian vaults, and any folder of plain text. Apple Notes support is coming soon.
This is intentional. Markdown is the simplest thing that works—readable in any editor, portable across tools, durable across decades. No database, no proprietary sync, no lock-in. Your Readwise highlights, your quick captures, your evergreen notes—all just text files in folders.
For PDFs and other documents, use docling to convert to markdown first.
Do I need to organize first? ▼
No. Incomplete systems still remember. Your half-finished tags, your links that go nowhere—Enzyme treats them as intent, not failure. Even scattered captures reveal the threads between them. That's enough.
What gets sent to the cloud? ▼
During enzyme init, excerpts are sent to AI for analysis. Your raw files stay on your device. Nothing is stored on remote servers.
Go deeper
enzyme petri | jq '.petri_entities[] | .entity' See what entities Enzyme found, check activity levels, explore catalysts
Questions tuned to what's latent in each cluster. They help you recognize which threads are yours—not by matching words, but by speaking to the same question.
Shell scripts to help describe your vault structure before running init