Stop Full‑Text Searching: Turn Any Project into a Queryable Knowledge Graph with Graphify

What is graphify

graphify maps a whole project (code, documentation, PDFs, images, videos) into a traversable knowledge graph, enabling queries, relationship lookup, and path tracing without opening files or performing full‑text search.

Core design

Local code parsing : Uses tree‑sitter to build an AST on the developer’s machine, guaranteeing deterministic extraction with no code upload.

Edge provenance : Each edge is labelled EXTRACTED (explicitly present in source) or INFERRED (deduced).

Non‑vector graph : The result is a true graph structure, not an embedding store.

Output artifacts

graphify-out/
├── graph.html      # open in a browser, click nodes, search, filter by community
├── GRAPH_REPORT.md # highlighted summary: God nodes, surprising connections, suggested queries
└── graph.json      # full graph for programmatic queries

Typical workflow

Scan the project → generate the graph → ask for paths or explain nodes → modify code.

Common internal use cases: onboarding new projects, checking dependencies before bug fixes, aligning code reviews, and unifying documentation with code.

Installation

Python 3.10+ is required. Recommended installation via uv tool install graphifyy or pipx install graphifyy. After installing the CLI, register with AI assistants:

graphify install
graphify cursor install

In PowerShell use graphify . instead of /graphify to avoid path interpretation.

Key commands

explain : graphify explain "APIRouter" returns source file, community, degree, and upstream/downstream relations.

path : graphify path "FastAPI" "ModelField" returns the shortest connection, e.g.

FastAPI --uses--> DefaultPlaceholder <--references-- get_request_handler() --references--> ModelField

query : graphify query "what connects auth to the database?" returns a trimmed sub‑graph answering the natural‑language question.

GRAPH_REPORT.md sections

God Nodes : Highest‑degree concepts (e.g., Client/AsyncClient, Request/Response, HTTPTransport) useful for quick impact analysis.

Surprising Connections : Cross‑file or cross‑module edges marked INFERRED, indicating hidden coupling that should be reviewed.

Communities : Sub‑systems discovered by Leiden clustering, visualized as colored blocks to guide onboarding.

15‑minute hands‑on workflow for an unfamiliar repository

Generate the graph: graphify . (or /graphify . in an AI coding terminal).

Open GRAPH_REPORT.md and note God nodes related to the task (e.g., authentication, session, timeout).

Open graph.html in a browser, filter by community, and inspect suspicious nodes.

Query a path, e.g. graphify path "LoginHandler" "Timeout", to locate relevant code.

Modify the identified files. Optionally install the incremental hook with graphify hook install so future commits rebuild the graph automatically.

Local demo repository

The repository includes worked/httpx/ and worked/example/. Reproduce the example with:

/graphify ./worked/httpx/raw

Incremental rebuild hook

Commit graphify-out/ to version control. Installing the hook with graphify hook install runs an incremental rebuild after each commit; the AST extraction does not consume API quota.

Project URL

https://github.com/Graphify-Labs/graphify