Graphify-Labs/graphify

The problem

Coding agents burn context by grepping and reading files until they stumble onto the right answer. On a large repo this is slow, expensive, and unreliable, because the agent has no model of how modules, types and documents relate. Graphify parses a project into a graph and lets the agent query that graph instead of re-reading source. The persisted graph.json becomes a reusable artifact, so repeated questions don't re-scan the tree.

What's different

The usual response to this problem is a vector index: embed chunks, retrieve top-k. Graphify explicitly rejects that. Code is parsed with tree-sitter AST deterministically, with no LLM and no embeddings, and the output is a real graph you traverse. Two consequences set it apart. First, every edge is tagged EXTRACTED or INFERRED, so you can distinguish what was read from source from what was derived by resolution. Second, queries are structural (explain, path A B, scoped subgraphs) rather than similarity search, which suits call-chain and dependency questions that embeddings handle poorly. It also folds docs, PDFs, images and ADRs into the same graph, with a semantic pass that does use a model.

Who it's for

Developers using Claude Code, Cursor, Codex, Gemini CLI or similar assistants on medium-to-large codebases, especially ones with heavy cross-file coupling or a lot of design documentation. It is a good fit for teams worried about sending source to a hosted index. Anyone who mainly wants semantic search over prose, or who works in a small repo where grep is already fast, will not get much from it. Engineers wanting a pure RAG stack should look at vector-store tooling instead.

Cost of adoption

Requires Python 3.10+ and uv or pipx; installation is uv tool install graphifyy followed by graphify install to register the skill with your assistant. The code pass is local, but the docs/media pass calls your assistant's model or a configured API key, so privacy guarantees only fully hold for code. The output format carries a schema version, which is a maturity signal but also means integrations can break across releases. Apache-2.0. The README is heavily oriented around the commercial Graphify Cloud upsell, so watch for feature gating between the OSS CLI and the paid product.

Verdict

Graphify is young — created in April 2026 — yet already shows 134k stars and +6,875 in a single week, which looks more like a spike than a settled adoption curve. The core idea, deterministic AST graph plus explained edges, is genuinely different from embedding-based retrieval, and the zero-LLM code pass is a real cost and privacy advantage. The risk is two-fold: a star count that outruns a 203-day-old codebase, and a commercial roadmap that could pull capability behind the cloud product. Worth a trial on a large repo you already know well, so you can judge graph quality against your own understanding before depending on it.