Point knowledge-graph-basement at any folder of Markdown — an Obsidian vault, an LLM wiki, a digital garden — and get sixteen living views of what you know. One command, zero config, nothing leaves your machine.
Every Markdown vault quietly accumulates structure — wiki-links, clusters, hubs, orphans. Almost no tool surfaces it, so your knowledge base behaves like a warehouse: everything is in there, and you can't see any of it.
Search only finds what you already remember to look for. The connection between two ideas stays buried until you happen to open both files.
Pretty constellations you can't query, share, or act on. You zoom in, shrug, and close the tab. Nothing about your writing changes.
RAG chunks prose but throws away structure — how concepts link is exactly the context an agent needs to answer like you, not like a search engine.
The structure already exists — it's in your wiki-links. Nothing renders it usefully. Until now.
Any folder of .md — Obsidian vault, LLM wiki, digital garden, plain notes. No plugins, no front-matter ceremony.
The indexer extracts entities, links, sections and clusters at build time. Sixteen views all read the same index.
Output is 100% static. Drop it on Netlify, GitHub Pages, or a USB stick. Your notes stay exactly where they are.
The same graph, projected through different lenses — structural, relational, temporal, textual. Each one answers a question the others can't. Four ship today; the rest follow in MVP2 and MVP3.
Find your hubs, your orphans, and the threads you abandoned in March. The force graph and treemap show where your garden is dense — and where it's bare.
The index is plain JSON at build time: entities, links, clusters. Feed it as context so answers cite the structure of your knowledge, not vibes from a vector dump.
Make coverage, clusters and gaps visible in standup. Which topics are dense? Which haven't been touched since last quarter? Now it's a picture, not an argument.
Course notes in Markdown become a concept map: prerequisites, tangents, isolated islands. Fix the structure before your students find it.
| graph basement | Obsidian graph view | Gephi | grep / search | |
|---|---|---|---|---|
| Input | any folder of .md | Obsidian vault only | import plugins + tutorials | any text |
| Lenses on one index | 16 | 1 | 1 per manual session | 0 — a linear list |
| Publishes a shareable site | ✓ static, one URL | — | PNG export | — |
| Machine-readable index for your LLM | ✓ JSON at build time | — | — | chunks, no structure |
| Setup | 3 commands | already installed | steep | zero — but you outgrow it |
Product is pre-launch. Waitlist members get access in signup order — early-bird price is locked at signup.
No. Indexing and rendering are 100% local. The output is plain static files — you host them wherever you want, or don't host them at all. There is no account, no telemetry, no phoning home.
No — any Markdown folder works. The indexer reads wiki-links, frontmatter dates and headings. Obsidian vaults, LLM-wikis, digital gardens and plain note folders all work out of the box; no plugins to install.
The public demo runs on 226 real notes and the pipeline is comfortably fast there. The roadmap moves the index to SQLite + FTS5, with a design target of 10k+ documents. Huge corpora get incremental indexing first.
The core is MIT open source, forever — clone it and run it today at zero cost. The paid Cloud plan is for people who want auto-rebuild, custom domains and zero setup. Early-bird pricing is locked for waitlist members.
On the roadmap, deliberately after the visualization foundation: natural-language queries over your graph and a graph-of-thoughts view where the agent's reasoning chain becomes a visible whiteboard — Claude API with a local Ollama fallback.
Yes, if you're comfortable with git and npm — clone the repo and run three commands (see "How it works"). The four MVP1 views are live. Everyone else: join the waitlist and you'll get the hosted version the week it ships.
ONE LAUNCH EMAIL · NO SPAM · UNSUBSCRIBE IS ONE CLICK