You join a team. You're staring at 200,000 lines of code with no map. Understand Anything, a new open-source Claude Code plugin from developer Lum1104, wants to fix that. It uses a multi-agent pipeline to scan entire projects, mapping files, functions, classes, and dependencies into an interactive knowledge graph. It handles 26+ file types including Dockerfiles, Terraform, SQL, and Markdown, and exports to PNG, SVG, or filtered JSON.
Two views define the tool. The structural graph shows code architecture with clickable nodes. The domain view is more ambitious. It maps code to actual business processes using LLM-based relationship discovery. Instead of just seeing that auth.ts connects to user.ts, you see how those files serve the "user authentication" flow. There are also AI-generated guided tours that walk through the codebase in dependency order. A fuzzy and semantic search handles queries like "which parts handle auth?" And diff impact analysis shows ripple effects before you commit.
The plugin works with Claude Code and Cursor, GitHub Copilot, Gemini CLI, and OpenAI's Codex. That's not an accident. Most teams don't standardize on one AI coding tool, and Understand Anything meets that reality. Installation on Claude Code is two commands from the plugin marketplace. The project is MIT-licensed with a live demo. It also processes Karpathy-pattern LLM wikis, turning documentation into force-directed knowledge graphs with community clustering.
Hacker News commenters raised fair questions about whether this beats just asking Claude or Codex directly. Some compared it to Obsidian's graph view, pretty but maybe cumbersome in practice. The real test is whether teams actually use visual graphs for onboarding, or if they just ask agents questions and read the answers. The domain view and guided tours are genuine differentiators. But they're also the features most likely to feel like overhead when deadlines are tight.