👇 Live Demo Access / 在线体验 👇

   

⚠️ Public Demo Limitations: Hosted instances use shared API quotas. If you encounter rate limits (403/429), please deploy locally for the best experience.
⚠️ 演示环境说明: 中国用户请使用 SEOUL SERVER。在线服务使用共享 API 配额。如遇请求受限或响应缓慢，强烈建议 Clone 本地运行 以获取无限制的极速体验。

RepoReaper is an advanced, autonomous system designed for automated architectural analysis and semantic code search, transforming the typical "Chat with Code" experience. This innovative project implements an intelligent Agent that simulates the analytical skills of a Senior Tech Lead, enhancing the process of code understanding and research.

Instead of merely indexing repositories, RepoReaper leverages a Large Language Model (LLM) as the processing unit and integrates a Vector Store as a high-speed Context Cache. This dynamic approach enables the agent to efficiently navigate the repository structure, proactively pre-fetching critical information to ensure that users receive relevant insights tailored to their queries in real-time.

Core Philosophy: RAG as an Intelligent Cache

In typical code analysis tools, the concept of Retrieval-Augmented Generation (RAG) often functions as a static lookup mechanism. RepoReaper redefines RAG as a Dynamic L2 Cache for the LLM, which includes:

1. Repo Map Creation: Initially, the agent parses the Abstract Syntax Tree (AST) of the repository to create a lightweight symbol map.
2. Intelligent Prefetching: The initial analysis autonomously identifies key files based on architectural significance and populates the context cache (Vector Store).
3. Just-In-Time Readings: During user interactions, if the retrieval engine fails to deliver sufficient context, the system dynamically fetches relevant files from the repository, continuously updating the cache to refine responses.

Key Features:

AST-Aware Semantic Chunking

RepoReaper employs Python's ast module to implement Structure-Aware Chunking, maintaining code logic integrity and providing enriched context to the LLM.

Asynchronous Concurrency Pipeline

Designed for optimal performance, RepoReaper implements an asynchronous architecture facilitating high-throughput I/O operations and preventing bottlenecks during repository analysis and embedded context generation.

The "Just-In-Time" ReAct Agent

This sophisticated Reasoning + Acting (ReAct) agent refines user queries and engages with the codebase actively, ensuring accurate and context-aware responses without generating misleading information.

Hybrid Search Mechanism

RepoReaper integrates dense and sparse retrieval mechanisms, enhancing accuracy in responses by effectively combining semantic comprehension with exact keyword matching.

Native Bilingual Support

Optimized for English and Chinese environments, this architecture ensures seamless interaction regardless of the user's language preference through dynamic prompt engineering and a user-friendly interface.

Technical Stack:

• Core Technologies: Python 3.10+, FastAPI, AsyncIO
• LLM Integration: OpenAI SDK
• Vector Database: ChromaDB
• Search Algorithms: BM25Okapi, Rank-BM25
• Parsing Technology: Python ast
• Frontend Technologies: HTML5
• Deployment Options: Docker, Gunicorn, Uvicorn.

Performance Optimization

RepoReaper employs robust session management and efficient memory handling strategies to ensure a responsive and smooth user experience even under heavy loads.

RepoReaper represents a significant step forward in automated code auditing and analysis, offering powerful tools for developers and teams seeking to optimize their coding practices and enhance productivity.