TL;DR
Open-source tool indexes codebases into knowledge graphs and exposes them via MCP, enabling AI agents to understand architecture and avoid blind edits.
Key Points
- Supports 8 languages (TypeScript, Python, Java, C, C++, C#, Go, Rust) with AST-based parsing and cross-file import resolution
- MCP server integration works with Cursor, Claude Code, Cline, and other AI editors without per-project config
- Precomputes relational intelligence (clustering, call chains, execution flows) so agents get complete context in single tool call vs. multi-query chains
- Web UI runs entirely in WebAssembly (Tree-sitter, KuzuDB, in-browser embeddings) with zero server uploads; CLI indexes locally with gitignored .gitnexus/ storage
Why It Matters
AI coding assistants currently lack deep architectural awareness, leading to missed dependencies and broken call chains. GitNexus solves this by precomputing codebase structure as a queryable knowledge graph, enabling smaller models to compete with larger ones while reducing token overhead and improving reliability—critical for teams using AI agents in production codebases.
Source: github.com