How does the two-stage retrieval work?

The system first performs a high-speed vector search to identify the top 20 most relevant candidates. It then uses an LLM to rerank those candidates based on deeper context, delivering the top 5 most accurate results to the user.

What makes CCG-RAG different from standard grep search?

Unlike grep, which matches literal text strings, CCG-RAG uses semantic embeddings to understand the intent and functionality of your query, finding relevant code even if the variable or function names don't match your query terms exactly.

Does it support multiple programming languages?

Yes, it utilizes tree-sitter parsing which supports a wide range of languages including TypeScript and JavaScript, ensuring accurate code chunking and preservation of logical boundaries like classes and functions.

When should I rebuild the index?

It is recommended to rebuild the index after major architectural changes, large refactors, or significant PR merges to ensure that the semantic search results and knowledge graph reflect the current state of the codebase.

CCG-RAG Semantic Code Search

Name: CCG-RAG Semantic Code Search
Author: phuongrealmax

byphuongrealmax

•

開発者ツール

Enables deep codebase understanding and semantic search using advanced code embeddings and knowledge graphs.

概要

CCG-RAG (Code Guardian RAG) is a sophisticated search and retrieval engine that allows developers to interact with large repositories through semantic queries rather than simple text matching. By leveraging tree-sitter parsing for accurate code chunking and high-performance embeddings, it helps users find specific functionality, identify similar code patterns, and navigate complex documentation. It is particularly effective for onboarding to unfamiliar codebases or building context for complex refactoring tasks within the Claude Code environment, ensuring that every AI-driven change is informed by the existing architecture.

主な機能

Knowledge graph mapping of function calls and dependency hierarchies
Two-stage retrieval process with LLM-based reranking for high precision
Semantic code search across the entire codebase using vector embeddings
1 GitHub stars
Pattern-based retrieval to find similar implementations and conventions
Intelligent documentation search for APIs, guides, and architecture specs

ユースケース

Onboarding to a new or complex repository by exploring functionality semantically
Finding specific logic patterns or error handlers without knowing exact text matches
Building a comprehensive context map for large-scale code refactoring and impact analysis

概要

主な機能

Knowledge graph mapping of function calls and dependency hierarchies
Two-stage retrieval process with LLM-based reranking for high precision
Semantic code search across the entire codebase using vector embeddings
1 GitHub stars
Pattern-based retrieval to find similar implementations and conventions
Intelligent documentation search for APIs, guides, and architecture specs

ユースケース

Onboarding to a new or complex repository by exploring functionality semantically
Finding specific logic patterns or error handlers without knowing exact text matches
Building a comprehensive context map for large-scale code refactoring and impact analysis