harshkedia177/axon

What it solves

Axon creates a structural knowledge graph of a codebase to prevent breaking changes and reduce the cognitive load on developers and AI agents. It solves the problem of AI agents working with "flat text" (grepping for strings), which often leads to missed indirect dependencies and a lack of understanding of how code is actually connected.

How it works

Axon uses a 12-phase analysis pipeline that runs locally to index a repository. It extracts ASTs via tree-sitter, resolves imports, traces function calls, and analyzes type relationships. It then applies the Leiden algorithm for community detection and analyzes git history to find "change coupling" (files that frequently change together). This data is stored in a graph, which can be queried via a web dashboard, a CLI, or an MCP (Model Context Protocol) server that allows AI agents to perform hybrid searches (BM25 + Vector + Fuzzy) and impact analysis.

Who it’s for

• AI Agent Developers: Those wanting to give agents like Claude Code or Cursor deep architectural awareness without stuffing the context window.
• Software Engineers: Developers who need to visualize call graphs, detect dead code, or perform blast-radius analysis before refactoring.

Highlights

• Interactive Web UI: A force-directed graph visualization with community hull overlays and a Cypher query console.
• MCP Integration: Exposes tools like axon_impact and axon_context so agents can understand dependencies in a single tool call.
• Deep Analysis: Includes framework-aware dead code detection and execution flow tracing (e.g., detecting FastAPI or Express routes).
• Git Coupling: Identifies hidden dependencies by analyzing which files are co-changed in git history.
• Local-First: All parsing, embeddings, and graph storage run locally with zero cloud dependencies.