Knowledge graphs

Graph model

The knowledge graph is a directed graph stored in SQLite and serialized to .genome/graph.json.

• File nodes — one per source file (file:path/to/module.py)
• Symbol nodes — functions, classes, methods with qualified names
• Import edges — module dependencies
• Call edges — resolved invocations where possible
• Inheritance edges — class hierarchies

Incremental builds

After the first full scan, Codegenome can update only changed files. The scan cache at .genome/scan_cache.db avoids re-parsing unchanged content.

Use codegenome analyze . for a standard incremental build, or the legacy CLI:

python -m codegenome --workspace . --build
python -m codegenome --workspace . --build --full

Graph intelligence

Built-in analysis includes:

• Dead code detection
• Entry points
• God nodes (high betweenness)
• Circular dependencies
• Complexity and churn rankings
• Community detection (Leiden clustering on macro graphs)

These power MCP tools like get_dead_code and get_god_nodes. See MCP tools.

Timeline snapshots

Each build can append a snapshot to .genome/watcher.db, enabling:

• Timeline dumps per node
• Diffs between snapshot IDs
• Churn rankings over time

Query via legacy CLI flags: --dump-timeline, --dump-changes, --dump-churn.

Artifacts

All paths are relative to your workspace root (the project you analyze). Details: Workspace artifacts.