index_graph

Build a knowledge graph of a codebase using AST analysis

Overview

Walks a project directory and extracts a knowledge graph using tree-sitter AST analysis — no LLM required. Nodes represent structural units (files, functions, classes, methods); edges represent relationships (contains, calls, imports, inherits).

The graph is persisted as graph.json alongside the vector index and is used by search_graph and graph_neighbors. It is also blended into results by hybrid_search when present.

Parameters

Parameter	Type	Default	Description
path	string	required	Absolute path to the codebase root directory
force	boolean	false	If true, rebuilds the graph even if one already exists

Example

index_graph("/path/to/myproject")
→ "Graph built for /path/to/myproject: 496 nodes, 1251 edges, 35 files processed."

Rebuild after a major refactor:

index_graph("/path/to/myproject", force=True)

Node kinds

Kind	Description
file	A source file
function	A top-level or module-level function
class	A class definition
method	A method inside a class

Edge kinds

Kind	Description
contains	A file or class contains a child node
calls	A function or method calls another
imports	A file imports another module
inherits	A class inherits from another

Notes

• Graph is stored at ~/.vecgrep/<project_hash>/graph.json
• Independent of the vector index — can be run before or after index_codebase
• Runs entirely locally — no network calls, no LLM
• hybrid_search degrades gracefully to pure vector search if no graph index exists
• Use force=True after significant refactors to rebuild from scratch