Knowledge Graph

At the heart of noodlbox is a knowledge graph - a structured representation of your code that captures symbols, relationships, and organizational patterns.

Why a Graph?

Code is inherently relational:

• Functions call other functions
• Classes inherit from other classes
• Modules import other modules
• Features span multiple files

Relational databases struggle with these interconnections. A graph database makes traversing relationships natural and fast.

Node Types

CodeSymbol

The fundamental unit: functions, classes, methods, variables, interfaces, enums, and more.

{
  name: "authenticateUser",
  symbol_type: "Function",
  file_path: "src/auth/authenticate.ts",
  line_number: 42,
  content: "export function authenticateUser(...) { ... }",
  signature: "authenticateUser(credentials: Credentials): Promise<User>"
}

Symbol types by language:

File

Source files in your repository.

{
  path: "src/auth/authenticate.ts",
  extension: "ts"
}

Relationship Types

CALLS

Function/method call relationships.

(authenticate:CodeSymbol)-[:CALLS]->(validateToken:CodeSymbol)

CONTAINED_BY

Symbols contained in files.

(authenticate:CodeSymbol)-[:CONTAINED_BY]->(authFile:File)

Querying the Graph

noodlbox uses Cypher, a declarative graph query language:

Find all callers of a function

MATCH (caller:CodeSymbol)-[:CALLS]->(target:CodeSymbol)
WHERE target.name = "authenticateUser"
RETURN caller.name, caller.file_path

Storage Architecture

noodlbox uses LanceDB as a unified storage layer for both graph queries and full-text search.

Graph Storage

Graph relationships are stored and queried using lance-graph, enabling:

• Fast traversals across millions of relationships
• Complex pattern matching with Cypher queries
• Efficient aggregations

Full-Text Search

Code search uses LanceDB's built-in FTS with BM25 ranking (powered by Tantivy). This provides:

• Relevance-ranked search results
• Code-aware stopword filtering
• Fast incremental indexing

Learn More

• MCP Cypher Query Tool - Execute graph queries