@fengrru/codegraph

In-memory heterogeneous code graph with PageRank centrality ranking. Build a knowledge graph from your codebase and search it with multi-signal fusion.

Install

npm install @fengrru/codegraph
# Optional: npm install web-tree-sitter (for AST parsing)

Quick Start

import { CodeGraph, CodeGraphSearcher, CodeGraphRanker } from "@fengrru/codegraph"

// Build graph from codebase
const graph = new CodeGraph()
await graph.build("./src")

// Search
const searcher = new CodeGraphSearcher(graph)
const results = searcher.search("authenticateUser", { maxResults: 10 })

// Rank by PageRank
const ranker = new CodeGraphRanker(graph)
const ranked = ranker.rank()

Features

• 3 node types: file, symbol, module
• 9 edge types: contains, imports, calls, extends, implements, etc.
• PageRank centrality: d=0.85, maxIter=100
• k-hop ego graph: BFS-based subgraph extraction
• Token estimation: for prompt injection
• Incremental updates: file watcher with hot reload

Modules

| Module | Description | |--------|-------------| | CodeGraph | Core graph engine | | CodeGraphSearcher | Multi-mode search with relevance scoring | | CodeGraphRanker | PageRank + centrality ranking | | CodeGraphWatcher | Incremental file change detection |

Documentation

• API Reference — TypeDoc-generated API docs
• Source Code
• Examples

License

MIT