01 / 08

Tech Sharing · Pure Substance

Build a Personal Knowledge
Graph with Graphify

One command · fully multimodal · honest audit — turn any folder into a navigable knowledge network.

↑ The background is a real knowledge graph generated by Graphify

02 / 08
Part 01

Why Graph?

folder → tree → graph: the next step in human cognition

03 / 08

Feature Map

One tool, four jobs

📂

Folder Ingest

Recursively scans any path: md / pdf / code / images

🧠

Entity Extract

Uses an LLM to extract concepts, people, events and relations

🕸️

Force Graph

D3 force layout; click to jump to the source

🔍

Audit Trail

Every edge traces back to a source span

It's not "yet another RAG" — it draws the retrieval results so you see the shape of your information at a glance.

04 / 08

One command

From zero to graph in about 90 seconds

$ graphify ~/notes --out ./graph

# graphify.config.yaml
ingest:
  paths: [~/notes, ~/code/docs]
  include: ["*.md", "*.pdf", "*.py"]

extract:
  model: claude-opus-4-6
  schema: [concept, person, event, relation]

render:
  engine: d3-force
  audit: true     # every edge carries a source span
05 / 08

Efficiency Race

Without a knowledge base vs with one

Without
a KB
🛵
Feed it again… tidy up… forget again…
With
a KB
🏎️
AI finds it → confirm → get to work!
5×

Faster

-80%

Less re-feeding

∞

Persistent memory

06 / 08

Pipeline

End to end in 4 steps

📂
Scan
Read files recursively
→
🔬
Extract
LLM extracts entities
→
🕸️
Build
Build graph + dedupe
→
🎨
Render
D3 interactive graph

Every step has an audit log: you always know why a node exists and which line of which file it came from.

07 / 08

Try it tonight

Graphify your folders

$ npm i -g @lewis/graphify

$ graphify ~/obsidian-vault

#knowledge-graph #open-source #claude-agent #obsidian #d3-force
08 / 08
Thanks.

github.com/lewis/graphify · star / issue / PR welcome