Graphify: Turn Code Into a Queryable Knowledge Graph

Graphify is an open-source skill that converts any folder of code, databases, scripts, and documentation into a queryable knowledge graph. Instead of feeding your AI assistant raw files or token-heavy context windows, Graphify structures your codebase as a graph—so Claude Code, Cursor, Gemini CLI, and other tools can understand relationships between functions, schemas, dependencies, and docs at query time.

AI-generated content: drafted with generative AI (Claude, by Anthropic), fact-checked and approved by a human editor. Full notice at the end of the article.

Source repository Graphify-Labs/graphify ↗ 125,997 stars · Python

The problem: context bloat and lost relationships

When you ask Claude Code to navigate a large codebase—say, a full-stack app with hundreds of files, database schemas, infrastructure configs, and docs—you face two problems:

Context window limits. Dumping raw files into the chat burns tokens fast and often gets truncated.
Relationship blindness. Claude sees functions in isolation. It doesn't automatically know that UserService.create() relates to the users table, or that a config file controls retry behavior across three modules.

This forces you to manually guide Claude through the codebase or regenerate context for each question—slowing down code generation and debugging.

How Graphify works: parsing to graph

You type /graphify in your AI coding assistant and it maps your project into a knowledge graph you query instead of grepping files. From the README:

Code parsing: tree-sitter ASTs across ~40 languages resolve calls, imports and inherits links between files. Deterministic, no LLM, nothing leaves your machine.
Beyond code: docs, PDFs, images and video/audio also map into the graph; that semantic pass uses your assistant's model or a backend you configure.
Explained edges: each connection is tagged EXTRACTED (explicit in the source) or INFERRED (resolved by Graphify).
Not a vector index: no embeddings or vector store; a real graph you traverse.
Output: graph.html (clickable), GRAPH_REPORT.md (key concepts and suggested questions) and graph.json (the full graph).

Then you query it: graphify explain "APIRouter", graphify path "A" "B", or plain questions through your assistant.

Graphify in a Claude Code workflow

Graphify acts as a semantic context layer between you and Claude Code. Instead of pasting file paths and hoping Claude understands the connections, you:

Index your codebase once with Graphify (pointing it at your project root).
Ask Claude Code questions like: "Add a new user permission check to every authenticated endpoint" or "Refactor this database migration to handle the new schema without downtime."
Claude queries the graph, understands the dependency tree and schema relationships, and generates code that's already aware of your architecture.

This is especially valuable for large or unfamiliar codebases. Instead of spending time manually extracting "where are all the database queries?" or "which functions touch the user table?", Claude already knows—because the graph made those relationships explicit.

See also: Claude Code skills for other context-aware extensions.

Real limitations and when to skip it

Static snapshot: the graph reflects the code when you built it. Rebuild after big changes, and expect gaps where behaviour is decided at runtime (reflection, dependency injection, heavy metaprogramming).
INFERRED edges are guesses: the tags let you check them; verify before acting on one.
Docs and media cost tokens: code parsing is local and free, but the semantic pass over docs, PDFs and media uses a model.
Not a replacement for tests: a graph shows that a function is called, not that the call is correct.
Upsell: the README promotes Graphify Cloud (paid) for always-on and cross-repo graphs; the open-source CLI works on its own.

Skip it if your codebase is small enough that Claude Code navigates it comfortably with plain file reads and search.

Getting started

Install the CLI and register the skill
uv tool install graphifyy      # or: pipx install graphifyy
graphify install               # registers the skill with your AI assistant

Then, inside Claude Code:

Build the graph
/graphify .

Open graphify-out/graph.html to explore, read GRAPH_REPORT.md for the highlights, and ask Claude questions that the graph can answer. Note the package name: graphifyy, with two y's. Graphify is Apache-2.0 licensed and works in Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot and other assistants.