GitNexus

GitNexus, built by Akon Labs (YC S26), is an open-source code intelligence engine that transforms any codebase into a deterministic knowledge graph, then exposes it to AI agents through MCP tools. It is available as a free CLI/npm package, a no-install browser Web UI, and a managed Enterprise SaaS or self-hosted deployment.

What It Is

GitNexus sits between your source code and your AI coding agent. Instead of letting agents grep through files and guess at structure, GitNexus parses every file with Tree-sitter, resolves imports, call chains, field types, and return types across the entire codebase, and stores the result as a precomputed knowledge graph. Agents then query that graph through 17 MCP tools — getting exact callers, execution flows, and blast-radius analysis in a single compact response rather than burning tokens on repeated file exploration.

How the Indexing Pipeline Works

GitNexus builds its graph through a six-phase pipeline:

• Structure — walks the file tree and maps folder/file relationships
• Parsing — extracts functions, classes, methods, and interfaces using Tree-sitter ASTs across 17 languages (TypeScript, JavaScript, Python, Java, Kotlin, C#, Go, Rust, PHP, Ruby, Swift, C, C++, Dart, and more)
• Resolution — resolves imports, call chains, heritage, and constructor inference deterministically across files; user.address.getCity().save() resolves at every hop with zero embedding guesswork
• Clustering — Leiden community detection groups symbols into functional clusters (auth-core, billing, ingestion) scored by cohesion and modularity
• Processes — traces execution flows from entry points through call chains
• Search — builds hybrid BM25 + semantic + RRF indexes for fast retrieval

The graph is built once and reused forever. Auto-reindexing on commit keeps it current.

Agent Integration and MCP Tools

GitNexus exposes 17 MCP tools including query, context, impact, trace, detect_changes, rename, cypher, route_map, api_impact, and pdg_query. The impact tool returns blast-radius analysis grouped by depth with confidence scores; detect_changes maps a git diff to affected execution flows before a commit ships. Setup is a single command — npx gitnexus analyze indexes the repo, and npx gitnexus setup writes the MCP config for all detected editors automatically.

Supported editors include Claude Code, Cursor, Codex, Windsurf, Antigravity (Google), OpenCode, CodeBuddy (Tencent), and Qoder (Alibaba). Claude Code and Codex receive the deepest integration: MCP tools, agent skills, PreToolUse hooks that enrich searches with graph context, and PostToolUse hooks that detect a stale index after commits.

Benchmark Results

According to Akon Labs' published benchmark on DeepSWE (113 tasks, 89 projects), agents using GitNexus solved 68.4% of issues compared to 37.0% with no graph — a +31.4 point improvement. The vendor reports this was achieved on fewer tokens (−5.3% output tokens) and fewer steps (−10.6%). On the hardest 58 tasks, the solve rate was 55.9% versus 15.9% for the bare model. The vendor also reports a cost of $0.88 per solved task against $1.79 for the bare model.

Deployment Model

GitNexus runs in three modes:

• CLI + MCP — installs via npm (npm install -g gitnexus), indexes locally, serves all repos through one MCP server; everything stays on-machine with no network calls
• Web UI — runs entirely in the browser via WebAssembly at gitnexus.vercel.app; no install, no server, code never leaves the machine
• Docker — two signed images (CLI/server + Web UI) published to GHCR and Docker Hub, deployable via docker compose up -d; also one-click deployable to Render

Enterprise adds managed SaaS, self-hosted VPC deployment, and air-gapped operation with zero external calls, SSO, access controls, automated PR review with blast-radius analysis, auto-reindexing, and multi-repo unified graphs.

Update: v1.6.10

The latest release is v1.6.10, published on 2026-08-27, according to the GitHub repository. The project has been actively developed since its creation in August 2025, accumulating over 45,000 GitHub stars and 5,000 forks. The vendor reports 60,000+ weekly npm downloads. Recent completed roadmap items include constructor-inferred type resolution, wiki generation, multi-file rename, git-diff impact analysis, process-grouped search, Claude Code hooks, multi-repo MCP, zero-config setup, and hybrid search with vector indexing. Active development targets LLM cluster enrichment, AST decorator detection, and incremental indexing.