Summary

GitNexus addresses the limitations of AI coding tools that lack deep codebase structural understanding. It provides a client-side, zero-server knowledge graph creator and Graph RAG agent, enabling users to analyze code dependencies and relationships interactively. This empowers developers to understand the impact of changes ("what breaks if I change this?") and receive AI-generated insights with verifiable citations, all while keeping code private within the browser.

How It Works

The engine operates entirely within the browser using WebAssembly, leveraging Tree-sitter for Abstract Syntax Tree (AST) parsing and KuzuDB for both graph database storage and native HNSW vector indexing. It employs a two-phase indexing process: first building a comprehensive code knowledge graph, then generating embeddings in the background. This client-side, zero-deployment-cost architecture facilitates advanced features like hybrid search (BM25, semantic, graph expansion) and relational analysis via Cypher queries.

Quick Start & Requirements

Clone the repository, install dependencies (npm install), and run the development server (npm run dev). Access the application at http://localhost:5173, then drag and drop a ZIP archive of your codebase. Currently supports TypeScript, JavaScript, and Python; Go, Java, and C are in progress.

Highlighted Details

• Full WASM Stack: All core components—parsing, database, embeddings—run client-side via WebAssembly.
• Hybrid Search: Integrates BM25, semantic vector search, and 1-hop graph expansion for context-rich queries.
• Grounded AI: AI-generated answers are directly linked to [[file:line]] proof within the codebase.
• Client-Side Privacy: Code is processed exclusively in the user's browser, with no server uploads.
• Advanced Analysis: Supports codebase-wide audits, blast radius analysis, dead code detection, and dependency tracing.

Maintenance & Community

The project is actively being developed, with a roadmap outlining future features like Model Context Protocol (MCP) support and external database integration. No specific community channels or contributor details are listed in the README.

Licensing & Compatibility

Released under the MIT License, which is permissive for commercial use and integration into closed-source projects.

Limitations & Caveats

Support for Go, Java, and C languages is currently in progress. The project has encountered and documented WASM memory limitations related to Copy-on-Write behavior in KuzuDB when handling large numbers of embeddings, necessitating careful memory management. While V2 significantly improves visualization performance over V1, extremely large codebases may still present performance challenges. Key features like MCP server exposure are still on the roadmap.