Codebase Search

Overview

Every developer knows the frustration of searching for code in a large codebase. You know the functionality exists, but you can’t remember the function name, the file path, or the exact string to grep for. Codebase Search solves this by letting you describe what you’re looking for in plain English.

The skill indexes your entire codebase using semantic embeddings, creating a searchable vector store that understands code meaning. When you search for “rate limiting middleware,” it finds your throttle.ts even if the word “rate” never appears in the code.

Built on LangChain with support for multiple vector stores (Chroma, Pinecone, FAISS), it handles repositories from hundreds to millions of lines of code.

How It Works

1. Index — The skill chunks your codebase into semantic units (functions, classes, modules) and generates embeddings
2. Store — Embeddings are stored in a local or remote vector database
3. Search — Natural language queries are embedded and matched against code semantics
4. Return — Results ranked by relevance with file paths, line numbers, and context

Use Cases

• Onboarding — New team members find relevant code without tribal knowledge
• Refactoring — Discover all code related to a concept before making changes
• Documentation — Find undocumented functionality by describing expected behavior
• Code reuse — Discover existing utilities instead of reinventing them
• Debugging — Trace functionality across complex codebases

Getting Started

# Install
pip install codebase-search

# Index your repo
codebase-search index --repo ./my-project

# Search
codebase-search find "email validation logic"

Example

$ codebase-search find "function that retries HTTP requests with exponential backoff"

Results:
1. src/utils/http.ts:42 - retryWithBackoff() [score: 0.94]
2. src/api/client.ts:118 - fetchWithRetry() [score: 0.87]
3. tests/utils/http.test.ts:15 - describe("retry logic") [score: 0.72]

Alternatives

• Sourcegraph — Code search and intelligence platform
• GitHub Code Search — GitHub’s built-in code search with regex support
• Bloop — AI-powered code search using natural language