Semantic Code Search

Find code by meaning, not just text. Search across codebases using natural language intent.

Quick Start

from code_search import CodeIndex, SemanticSearch

index = CodeIndex("/path/to/codebase")
index.build()  # Parse AST, generate embeddings

search = SemanticSearch(index)
results = search.query("how is authentication handled?")
results = search.similar_to("src/auth/login.py:validate_token")

How It Works

1. Parse — Walk codebase, extract functions/classes with AST
2. Embed — Generate vector embeddings from code + docstrings
3. Index — Store in vector index with metadata (file, line, type)
4. Search — Query by intent or find similar code

Search Types

• Intent search: "find error handling patterns" → returns matching code
• Similarity search: Given a function, find others doing the same thing
• Structural search: Find all functions matching a call pattern
• Duplicate detection: Find code doing the same thing differently

CLI

python3 scripts/search.py index /path/to/codebase
python3 scripts/search.py query "database connection setup"
python3 scripts/search.py similar src/db/connect.py:10
python3 scripts/search.py duplicates