How to install Graphify and build a code graph

What do you need before installing Graphify?

You need Python 3.10 or newer and an AI coding assistant that can load project skills. The documented setup uses Claude Code, while Codex, Cursor, and other assistants can consume the generated graph artifacts or expose the graph through MCP.

• Check the active Python version before installing so pip does not target an old interpreter.
• Use pipx when the operating system blocks global pip installs or when Python is externally managed.
• Open the assistant in the repository or folder you want to map; it needs permission to read that project.

How do you install Graphify with pip?

Run pip install graphifyy and then graphify install. The package name is a temporary PyPI workaround: graphifyy installs the software, while graphify remains the command you use.

• Python package managers may need a user-level install if the global environment is protected.
• On macOS systems with externally managed Python, pipx install graphifyy is the lower-friction path.
• After installation, restart the assistant or open a new session if the /graphify skill does not appear immediately.

How do you build the first codebase graph?

Open the target repository in your AI assistant and invoke /graphify followed by a path. Use /graphify . for the current project, or point it at a narrower folder when you want a smaller first map.

• Start with a representative subsystem if a full monorepo would take too long.
• Let the first build finish before asking architecture questions so every artifact shares the same snapshot.
• Inspect graph.html for the interactive view and GRAPH_REPORT.md for god nodes, surprising connections, and suggested questions.

How do you update a graph after the code changes?

Run the same graph workflow with --update to process changed files and merge them into the existing graph. For repositories under active work, --watch rebuilds code structure after file saves, while graphify hook install rebuilds after commits.

• Use update mode for controlled checkpoints after a batch of changes.
• Use watch mode when multiple agents or developers are editing the same repository in parallel.
• Use the Git hook when you want freshness tied to commits without keeping a background watcher running.

How do you query and reuse the graph?

Ask the assistant to query graph.json for structural questions, or use Graphify's query, path, and explain commands outside the assistant. The graph persists across sessions, so the next task can begin from the existing map instead of rereading every source file.

• Use query for open-ended relationships and entity connections.
• Use path when you need the route between two symbols or concepts.
• Use explain when you need the evidence and context around a single node.

What should you do if graphify is not recognized?

First confirm that the Python scripts directory is on PATH, then retry the command from a new shell. Windows commonly needs the user-level Python Scripts folder; pipx avoids most PATH issues by placing commands in its managed binary directory.

• Run the package installer's command listing to confirm that Graphify was installed for the expected interpreter.
• Restart the terminal and AI assistant after changing PATH.
• If the package installed successfully but the skill is missing, repeat graphify install and restart the assistant session.

Does Graphify send the whole repository to a model?

Graphify extracts code structure locally with tree-sitter and deterministic parsing. Semantic extraction for documents, PDFs, screenshots, and diagrams depends on the configured assistant model, so teams should review their model and data-handling policy before adding sensitive non-code material.

• Code graph construction is separated from the semantic model pass.
• Each edge is labeled EXTRACTED, INFERRED, or AMBIGUOUS so reviewers can see how a relationship entered the graph.
• Keep secrets and generated caches out of the input folder when you configure a new project.