GitNexus is a client-side code intelligence engine designed to enhance code exploration and analysis, particularly for AI agents. Its primary purpose is to provide a deep understanding of a codebase's structure and relationships, enabling more reliable and efficient interactions with code, especially for AI-powered tools like code completion and refactoring assistants. Unlike traditional approaches that rely on raw graph edges, GitNexus precomputes structural information, such as dependencies, call chains, and functional clusters, at index time. This precomputed relational intelligence allows AI agents to access complete context with a single query, leading to improved reliability, token efficiency, and model democratization.

The core functionality of GitNexus revolves around two primary modes of operation: the CLI + MCP (recommended) and the Web UI. The CLI, installed via `npm install -g gitnexus`, allows for local indexing of repositories. The `gitnexus analyze` command indexes a codebase, creating a knowledge graph that captures every dependency, call chain, and execution flow. This indexed information is then exposed through an MCP (Message Communication Protocol) server, enabling AI agents to access the codebase's structure. The CLI also provides commands for setting up MCP integration with various editors like Cursor, Claude Code, Windsurf, and OpenCode, as well as tools for managing the index, such as listing indexed repositories, checking index status, and cleaning the index. The CLI is designed for daily development and full-scale repo analysis.

The Web UI, accessible at [gitnexus.vercel.app](https://gitnexus.vercel.app), offers a browser-based, interactive graph explorer and AI chat interface. Users can drag and drop a ZIP file of a repository to begin exploring the code. The Web UI uses the same indexing pipeline as the CLI but operates entirely within the browser using WebAssembly (WASM). This mode is ideal for quick exploration, demos, and one-off analyses. It is limited by browser memory, but can be connected to the CLI's local backend server for access to larger repositories.

GitNexus's key features include a multi-phase indexing pipeline that encompasses structure mapping, parsing with Tree-sitter, dependency resolution, clustering, process tracing, and hybrid search. The CLI exposes seven tools via MCP, including tools for listing repositories, querying the codebase, providing 360-degree symbol views, performing impact analysis, detecting changes, renaming symbols, and executing Cypher graph queries. These tools provide AI agents with comprehensive context, enabling them to understand the impact of code changes, identify dependencies, and perform refactoring tasks more effectively. The CLI also provides resources for instant context, such as lists of indexed repositories, codebase statistics, and execution flows. Furthermore, GitNexus automatically installs agent skills for exploring code, debugging, impact analysis, and refactoring, enhancing the capabilities of AI-powered tools.

The project's architecture supports multi-repo indexing, allowing a single MCP server to serve multiple indexed repositories. The CLI stores the index within the repository's `.gitnexus/` directory, which is gitignored, and registers a pointer in a global registry. The MCP server reads this registry to serve any indexed repository. The Web UI and CLI are built using a modern tech stack, including Node.js, Tree-sitter, KuzuDB, and HuggingFace transformers.js. The project emphasizes security and privacy, with all CLI operations running locally and the Web UI operating entirely within the browser, ensuring that code never leaves the user's machine. The project is open source, allowing for independent code audits.