Map the whole project into a queryable knowledge graph instead of grepping
Graphify-Labs/graphify
YC S26, Apache-2.0, 120k stars, Python 3.10+. /graphify . maps code, docs, PDFs, images and videos into a graph and emits a clickable graph.html, a human-readable GRAPH_REPORT.md and a machine-readable graph.json. Code goes through tree-sitter AST parsing (about 40 languages): deterministic, no LLM calls, nothing leaves the machine.
Our takeThe design that matters most: every edge is labelled EXTRACTED or INFERRED, so you can tell what the source explicitly contains from what the tool inferred — which makes an agent's conclusions auditable, a hard requirement in production. It is not a vector index: no embeddings, no vector store, a real graph you can explain, path and query. In its own benchmark, building the graph costs zero LLM spend and ingest is an order of magnitude cheaper, a structural win from local parsing.
uv tool install graphifyy && graphify installIn one line: turn the project into a graph you query, not a tree you grep
Type /graphify . in your AI coding assistant and it maps the entire project — code, docs, PDFs, images, videos — into a knowledge graph you can query instead of grepping through files. 119,837 stars / 11,587 forks, Apache-2.0, YC S26, primary language Python (3.10+).
Three design decisions separate it from "yet another RAG layer", and the README states them bluntly:
- Code maps for free, fully local. Code is parsed with tree-sitter AST: deterministic, no LLM, nothing leaves your machine (~40 languages). Only docs, PDFs, images and video use your assistant's model or a configured API key for the semantic pass.
- Every edge is explained. Connections are tagged
EXTRACTED(explicit in the source) orINFERRED(resolved by graphify), so you can tell what was read from what was guessed. - Not a vector index. No embeddings, no vector store — a real graph you traverse. Ask a question, trace a path between two things, or explain one concept.
The second one matters most to us. Most knowledge indexes hand you a similarity score you cannot interrogate; the EXTRACTED/INFERRED split makes an agent's conclusions auditable, which in production is a hard requirement rather than a nicety.
Outputs and CLI
graphify-out/
├── graph.html # force-directed, clickable, with a community legend
├── GRAPH_REPORT.md # human-readable structure report
└── graph.json # machine-readable for downstream toolingThree main CLI verbs: explain (one concept), path (route between two nodes), query (a question). On the analysis side it finds god nodes (very high in-degree hubs, usually your architectural risk points) and runs Leiden community detection with LLM-free community labels. One detail we like: # NOTE: and # WHY: comments plus ADR references get promoted into graph nodes — meaning design intent becomes searchable structure, not just a symbol table.
Its own published benchmarks, with methodology
| Metric | graphify | Comparison |
|---|---|---|
| LOCOMO recall@10 | 0.497 | mem0 0.048 · BM25 0.362 |
| LOCOMO QA | 45.3% | — |
| LongMemEval-S | 76% | tied for best |
| LLM credits to build the graph | $0 | — |
| Ingest cost | $1.40 | supermemory $15.67 |
The methodology is published too (BENCHMARKS.md): one harness, Kimi K2.6, blind judges, inter-rater κ=0.81. This is vendor self-evaluation and we relay it as such rather than as independent verification — but at least the conditions are disclosed, which beats a bare bar chart. The interesting numbers are the last two rows: zero LLM cost to build, and ingest an order of magnitude cheaper. That is a structural consequence of local tree-sitter parsing, not tuning.
Install
uv tool install graphifyy && graphify installNote the PyPI name is graphifyy (double y). graphify install registers /graphify into your AI coding assistants; Claude Code, Codex, Cursor and Gemini CLI are supported.
The business model is stated plainly: the repo is usable under Apache-2.0, while the company pushes the hosted platform at app.graphify.com (early access open), whose selling point is being always-on and continuously updated across code, docs and meetings — whereas the CLI is on demand. Want always-on, use the platform; want local and free, use the CLI.