VECTRON

VECTRON maps your codebase into a live dependency graph. Click any node to simulate blast radius ,see what breaks before you break it. AI queries, MCP server, 5-agent analysis. Zero setup

Comment

Inspiration

Every developer knows the moment — you're about to refactor a critical function, but you have no idea what depends on it. You search manually, miss something, and the build breaks. Existing tools give you text and lists. Nothing visual. Nothing intelligent. VECTRON was built to fix that.

What it does

VECTRON transforms any codebase into a live, interactive knowledge graph. Paste a GitHub URL or upload a ZIP — VECTRON parses every file, function, class, and dependency using Babel AST and renders it in WebGL using Sigma.js with ForceAtlas2 layout.

Core features:

  • Blast Radius Simulation — Click any node, BFS propagation shows exactly what breaks if that file changes, color-coded by impact depth
  • AI Query Engine — Ask anything structural about your codebase, get answers with simultaneous graph node highlighting
  • Process Flow Detector — Auto-detects multi-step workflows and renders them as Mermaid diagrams
  • Intelligence Report — 5 AI agents run in parallel (Security, Architecture, Performance, Code Quality, Onboarding) and synthesize a unified report
  • MCP Server — 4 tools exposed via Model Context Protocol: Claude Code, Cursor, and Codex can query the live graph directly
  • Metrics Dashboard — Risk scores, complexity, circular dependency detection

How we built it

  • Frontend: React 18 + TypeScript + Vite, Sigma.js + Graphology + ForceAtlas2 for WebGL graph rendering
  • Backend: Node.js + Express, Babel AST for parsing JS/TS/JSX/TSX/Python
  • AI Layer: Groq (llama-3.3-70b-versatile) with Cerebras fallback
  • MCP Server: @modelcontextprotocol/sdk, SSE transport on port 3002
  • Diagrams: Mermaid.js for process flow visualization
  • Deployment: Railway

Challenges we ran into

  • Rendering 2000+ nodes at 60fps required careful WebGL optimization with Sigma.js and ForceAtlas2 layout tuning
  • GitHub ZIP download over fetch instead of git clone (Railway doesn't have git installed)
  • Building the BFS blast radius simulation to correctly traverse complex cyclic dependency graphs
  • Running 5 AI agents in parallel without hitting rate limits while keeping response times acceptable

Accomplishments that we're proud of

  • Blast radius simulation that visualizes cascading failures visually — something no other tool does
  • MCP server integration that gives AI coding assistants live structural context instead of hallucinated architecture
  • Zero setup — paste a GitHub URL and get a full knowledge graph in seconds
  • Handles real-world codebases with 2000+ nodes and 20000+ edges

What we learned

  • WebGL graph rendering at scale requires fundamentally different approaches than DOM-based visualization
  • Babel AST is incredibly powerful for extracting deep structural relationships across JS/TS codebases
  • MCP protocol unlocks a new category of AI-native developer tooling
  • Blast radius simulation is a genuinely unsolved problem in developer tooling

What's next for VECTRON

  • GitHub PR Risk Analyzer — automatically run blast radius on every changed file in a PR
  • Refactor Agent — plan → execute → verify loop using MCP tools
  • VS Code Extension — bring the graph directly into the editor
  • CI/CD Integration — block merges when blast radius exceeds risk threshold

Built With

Share this project:

Updates