A Markdown corpus Agent + multi-visualization playground. Take any folder of
.mdnotes → see it as 16 different graphs.
A tool that indexes a Markdown corpus (LLM-wiki / Obsidian vault / digital garden / plain notes folder), extracts the knowledge graph (entities + sections + links), and renders the corpus through 16 different visualizations — from classic force-directed graphs to 3D knowledge galaxies to graph-of-thoughts.
Phase 0 — scaffolding + corpus exploration. See docs/progress.md for current state.
./scripts/bootstrap.sh # one-time setup (clone test corpus, install deps)
npm run index # parse corpus → data/index.json
npm run stats # print graph statistics
npm run dev # start viz playground| Layer | Tech | Purpose |
|---|---|---|
| Parser | unified + remark + mdast-util-from-markdown |
Parse .md to AST |
| Indexer | Node.js (this repo) | Extract nodes, edges, sections, frontmatter |
| Storage | SQLite (FTS5) + JSON snapshots | Fast search + portable cache |
| Embeddings | sqlite-vec or LanceDB (later) | Semantic search |
| Agent | Claude API + local Ollama (later) | NL queries, summarization |
| Viz | Astro + D3.js + visx + Cytoscape + ECharts + regl | 16+ viz types |
| Deploy | Cloudflare Pages (later) | Static + edge |
See docs/viz-ideas.md for the full list and priority order.
This project uses caro401/notes as its initial test corpus (cloned into corpus/, gitignored). It's a real, populated Obsidian vault with ~228 single-concept notes, cross-linked, MIT-licensed. Swap in your own corpus by replacing corpus/.
Code: MIT. Test corpus (in corpus/) is MIT (caro401's license).