How to Use Graphify: Open‑Source AI‑Powered Software Engineering Knowledge Graph

Why AI Programming Needs a Knowledge Graph

Traditional software projects rely on READMEs, architecture docs, API docs, and code comments to share knowledge. In the era of AI assistants, these artifacts are still needed, but feeding them to a model requires repeatedly scanning files and stitching context, which becomes inefficient as projects grow.

Files are present, but understanding does not accumulate.

Graphify addresses this by extracting relationships among code, documentation, PDFs, images, and videos, producing a persistent, queryable knowledge graph that AI assistants can traverse.

What Graphify Does

When invoked, Graphify parses the project directory, performs local AST analysis on code (using tree‑sitter) and semantic extraction on unstructured assets via an AI model, then merges everything into a graph. The output consists of three files: graph.html – a browser‑viewable visualization of the graph. GRAPH_REPORT.md – a human‑readable summary highlighting key concepts, high‑degree nodes, surprising relationships, and suitable follow‑up questions. graph.json – the full structured graph for programmatic queries.

Each relationship is labeled with a source tag ( EXTRACTED, INFERRED, or AMBIGUOUS) and a confidence score, allowing users to distinguish factual links from model‑inferred ones.

Installation and Basic Commands

Graphify requires Python 3.10+ and the uv tool (recommended) or pipx. Install the core package with: uv tool install graphifyy or pipx install graphifyy Register the tool with your AI assistant: graphify install Generate a graph for the current project: /graphify . On Windows PowerShell omit the leading slash: graphify . Supported AI back‑ends include Claude Code, Codex, OpenCode, Cursor, Gemini CLI, GitHub Copilot CLI, VS Code Copilot Chat, Aider, Kimi Code, Kiro, Trae, and others. Platform‑specific commands (e.g., graphify install --platform windows) let you tailor the integration.

Querying the Graph

After the graph is built, use the built‑in commands to explore relationships:

/graphify query "what connects auth to the database?"

/graphify path "UserService" "DatabasePool"

/graphify explain "RateLimiter"

These queries go beyond simple grep by following the semantic links between modules, concepts, and external resources.

Incremental Updates and Advanced Usage

When the project changes, you can update only modified files: /graphify ./docs --update To rebuild only the clustering step without re‑extracting content: /graphify . --cluster-only To skip HTML visualization for large projects: /graphify . --no-viz Export a Markdown wiki: /graphify . --wiki Add external resources (papers, YouTube videos) directly into the graph:

/graphify add https://arxiv.org/abs/1706.03762

/graphify add <youtube-url>

Extras and Optional Features

Graphify offers optional extras that can be installed on demand, such as PDF extraction, Office file support, Google Sheets rendering, video/audio transcription, Neo4j export, SVG export, local Ollama inference, OpenAI, Gemini, Bedrock, and SQL schema extraction. Install an extra with, for example:

pip install "graphifyy[pdf]"

Team Collaboration and Best Practices

Commit the graphify-out/ directory to Git so teammates can reuse the generated graph without re‑running the full pipeline. Use a .graphifyignore file (syntax similar to .gitignore) to exclude noisy or irrelevant files. A typical workflow:

Run /graphify . locally.

Commit graphify-out/ to the repository.

Other developers pull the repo and the AI assistant reads the pre‑built graph.

Future changes are applied with --update or via a Git hook ( graphify hook install) that rebuilds the graph on each commit.

When merging graphs from multiple sources, use graphify merge-graphs a.json b.json. Remember that inferred or ambiguous edges are hints, not definitive conclusions, and should be reviewed manually.

Privacy, Cost, and Deployment Considerations

Code AST parsing stays on the local machine, and video/audio can be transcribed with faster‑whisper locally. However, unstructured assets (PDFs, images) are sent to the configured AI model for semantic extraction, which may incur API costs and raise data‑privacy concerns. For enterprise or sensitive projects, define clear boundaries for what data is sent to external services.

Graphify can run headlessly in CI pipelines, with environment variables selecting the backend (e.g., ANTHROPIC_API_KEY, GEMINI_API_KEY, OPENAI_API_KEY, OLLAMA_BASE_URL, AWS_* for Bedrock).

Conclusion

Graphify transforms scattered project artifacts into a structured, queryable knowledge graph, enabling AI assistants to follow relationships rather than repeatedly re‑scanning files. By persisting this “project memory,” developers gain faster context retrieval, more reliable AI‑driven assistance, and a shareable source of truth for teams.