GitNexus is a code knowledge graph tool for AI agents, designed to address the shortcomings of traditional grep-style code retrieval. When AI agents only view code file by file, they easily miss dependencies, break call chains, and make blind modifications. GitNexus indexes the codebase into a knowledge graph, allowing agents to read along resolved execution flows rather than searching file by file.
Main capabilities
- Knowledge graph indexing: Index any codebase into a knowledge graph containing every dependency, call chain, cluster, and execution flow.
- Compiler-level static analysis: Use Tree-sitter to parse every file, deterministically resolving imports, call chains, field types, and return types across the codebase, without embedding guesses.
- Architecture discovery: Group symbols into functional clusters via Leiden community detection, and score cohesion and modularity to automatically discover architectures such as authentication flows, billing engines, and ingestion pipelines.
- Change impact analysis: After modifying a function, trace all downstream callers, group them by depth and attach confidence scores; detect_changes can map a git diff to affected execution flows, knowing the impact scope before committing.
- Cross-repository impact analysis: Group multiple repositories into a unified graph, with cross-repository edges connecting APIs to their consumers, so that a breaking change in one service becomes visible across all dependent repositories before release.
- Unified query: Agents can query hundreds of repositories as a single graph, obtaining flow-level answers (callers, execution flows, impact scope), replacing large file dumps with compact lines, providing more accurate context and significantly reducing token consumption.
Usage GitNexus supports self-hosting and is suitable for enterprise deployment. It can be used with various editors and agents. An official graph browser is provided to query every cluster, call chain, and execution flow in real time on real codebases. Specific installation commands and configuration steps are not provided yet.
Use cases
- AI-assisted development of large codebases, helping agents understand system-level dependencies rather than local files.
- Change impact analysis in multi-repository microservice architectures, knowing the impact on other services before modifying one repository.
- Enterprise-grade code intelligence and self-hosting needs.
- AI programming workflows that need to reduce token consumption while improving context accuracy.
Limitations and notes GitNexus performs well on the DeepSWE benchmark: reading by flow alone can fix 68.4% of issues, versus 37.0% without the graph, while using fewer tokens and steps. However, the official statement pointsout that single-issue benchmarks cannot measure GitNexus's true purposes, such as unified graph queries across hundreds of repositories, impact scope analysis, taint tracking, and more. GitNexus is an open-source project and is one of the fastest-growing code intelligence projects on GitHub. Information such as supported languages, the specific list of editors and agents, enterprise edition feature details, and whether commercial use is permitted is not provided yet.
