DEV Community

Cover image for Graphify: Turning Your Notes Into a Knowledge Graph AI Agents Can Actually Query
Ansari Mohammed Bilal
Ansari Mohammed Bilal

Posted on • Originally published at ansaribilal.com

Graphify: Turning Your Notes Into a Knowledge Graph AI Agents Can Actually Query

7-minute read · October 2026

I have ~1,400 Markdown notes accumulated over 6 years — Obsidian, then Notion, then back to plain files. RAG on top of them works, sort of. What actually works is Graphify, an AI-agent skill that extracts entities and relationships from every note and hands the agent a real graph to traverse.

Graphify knowledge graph — glowing amber nodes in an isometric cube

RAG Isn't Enough (For Notes)

Semantic search on notes gives you the chunk that sounds most like the question. But my notes are dense with proper nouns — people, projects, tools, concepts. I don't want the chunk that sounds relevant; I want "every note that mentions Sanjay AND the Refero project." That's a graph query, not a vector one.

What Graphify Does

  1. Walks a folder of Markdown.
  2. Runs each file through an extraction pass — entities (Person, Project, Tool, Concept) and relationships (Sanjay WORKED_ON Refero, Refero USES Tailwind).
  3. Writes it all to a lightweight graph store (SQLite by default; Neo4j if you want).
  4. Exposes an MCP tool the agent can call: graph.query({ entities, relationships }).

Setup 

git clone https://github.com/... graphify
cd graphify && npm install
npm run index -- ~/notes
Enter fullscreen mode Exit fullscreen mode

Add to Claude Code:

{
  "mcpServers": {
    "graphify": {
      "command": "node",
      "args": ["/path/to/graphify/mcp.js", "--db", "~/notes/graph.db"]
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Ask Claude: "Who has worked on projects using Tailwind, based on my notes?" — you'll get a real answer, not a hallucination.

Where It Shines

  • Research synthesis. Ask "connect these three concepts" and the agent traverses actual edges.
  • CRM-lite. Track people, projects, and last-touched dates without a real CRM.
  • Post ideas. Ask "which topics have I mentioned >5 times and never blogged about?" — free content pipeline.

Where It Struggles

  • Entity resolution is noisy. "Sanjay" and "S. Kumar" get treated as different people until you add aliases.
  • Extraction cost. First index of 1,400 notes cost me ~$3 in Claude tokens. Incremental updates are free.
  • Schema drift. If you rename an entity type mid-way, re-index the whole folder.

Pair With

Bottom Line

RAG is a hammer. Graphify gives your agent a screwdriver too. For anything where the relationships matter more than the text — research, second brains, personal CRM — this is the shape of AI-native tooling I want more of in 2026.

I'm building a variant that indexes Twitter bookmarks. Follow along on X.

Originally published at ansaribilal.com. I write about AI agents, indie builds and developer tooling — more posts here.

Top comments (0)