Skip to content
On this page

    Understand Anything: Codebase Knowledge Graphs

    2 min read

    Today I discovered Understand Anything, a Claude Code plugin that solves the cold-start problem when joining a new codebase. You run one command and it builds an interactive knowledge graph of every file, function, class, and dependency in the project.

    The Problem It Solves

    You join a team. The codebase is 200,000 lines. You spend the first two weeks doing grep, reading stale docs, and asking teammates questions they’ve answered before. The mental model forms slowly and incompletely.

    Understand Anything automates the archaeological phase. It runs a multi-agent pipeline (5 specialized agents working in parallel) that combines tree-sitter for deterministic parsing with LLM analysis for semantic understanding. The structural side gives you accurate import/export graphs. The semantic side gives you plain-English explanations of what each file actually does.

    Setup

    # Install the plugin in Claude Code
    /plugin marketplace add Lum1104/Understand-Anything
    /plugin install understand-anything
    
    # Analyze your codebase
    /understand
    
    # Open the interactive dashboard
    /understand-dashboard

    The graph gets saved to .understand-anything/knowledge-graph.json. Commit it and teammates skip the pipeline entirely.

    Key Commands

    CommandWhat it does
    /understandBuild the knowledge graph (incremental by default)
    /understand-dashboardOpen interactive visualization
    /understand-chat <question>Ask questions about the codebase
    /understand-diffSee impact of your current changes
    /understand-explain src/file.tsDeep-dive into a specific file
    /understand-onboardGenerate an onboarding guide
    /understand-domainExtract business domain mapping

    The dashboard is fully interactive. You can pan, zoom, click nodes, and search across the graph visually. Try the live demo to see it in action before installing.

    What Makes It Interesting

    The tree-sitter + LLM hybrid approach is smart. Tree-sitter handles the deterministic stuff (imports, exports, call sites, inheritance) so the graph is reproducible. LLMs handle what parsers cannot: summaries, architectural layer assignments, business-domain mapping, guided tours.

    It also works across platforms. Claude Code natively, plus Cursor, Copilot, Codex, Gemini CLI, and others via a one-line install script.

    For monorepos, you can scope it to a subdirectory:

    /understand src/frontend

    And for ongoing use, --auto-update adds a post-commit hook that incrementally patches the graph with each commit.

    What I Learned

    • Knowledge graphs beat flat documentation for codebase comprehension because you can navigate relationships, not just read descriptions
    • The tree-sitter + LLM split is a pattern worth noting: deterministic parsing for structure, LLM for semantics
    • Committing the graph JSON means new team members get instant understanding without running the pipeline
    • Incremental updates keep it practical for daily use. Only changed files get re-analyzed
    • Works with Claude Code, Cursor, Copilot, Gemini CLI, and 10+ other platforms

    Onboarding to new codebases frequently? I’d love to hear what tools you’re using. Reach out on LinkedIn.