Egonex-AI/Understand-Anything

Graphs that teach > graphs that impress. Turn any code into an interactive knowledge graph you can explore, search, and ask questions about. Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI, and more.

📚 Understand Anything – Turn any codebase, knowledge base, or docs into an interactive knowledge graph

What it is – A plugin/CLI that runs on a variety of AI‑coding platforms (Claude Code, Codex, Cursor, VS Code + Copilot, Gemini CLI, etc.). It scans a project, extracts every file, function, class and dependency, then builds a JSON knowledge‑graph. The graph can be explored in a web dashboard where nodes are searchable, clickable, and annotated with plain‑English summaries, architectural layers, business‑domain mappings, and guided tours.

🎯 Core capabilities

Feature	What you get
Structural graph	Every source file, function and class becomes a node; edges show imports, calls, inheritance, etc.
Semantic enrichment	LLM agents add human‑readable descriptions, layer tags (API, Service, Data, UI, Utility), and business‑logic mapping.
Guided tours	Auto‑generated walkthroughs that order the code by dependency so you can learn the architecture step‑by‑step.
Fuzzy & semantic search	Ask “which parts handle auth?” and the system returns relevant nodes across the graph.
Diff impact analysis	After a change, the tool highlights which parts of the graph are affected.
Persona‑adaptive UI	Dashboard simplifies or expands details for junior devs, PMs, or power users.
Knowledge‑base mode	Point the tool at a markdown wiki (e.g., Karpathy‑pattern LLM wiki) and get a force‑directed graph of concepts, entities and claims.
Incremental updates	Only files that changed since the last run are re‑analyzed, saving tokens and time.
Multi‑language UI	Node summaries and UI can be generated in English, Chinese (simplified/traditional), Japanese, Korean, Russian.

🛠️ How it works under the hood

Tree‑sitter parser – Deterministically builds a concrete syntax tree for each file, extracts imports, definitions, call sites, and creates a stable importMap.
LLM agents – Take the parsed structure plus the raw source and produce natural‑language summaries, layer assignments, domain mapping, and tour scripts.
Multi‑agent pipeline – Five core agents (project‑scanner, file‑analyzer, architecture‑analyzer, tour‑builder, graph‑reviewer) run in parallel (up to 5 workers). Optional domain‑analyzer and article‑analyzer handle business‑process extraction and wiki parsing.
Output – A reproducible structural graph (same code → same edges) plus semantic annotations stored in .ua/knowledge-graph.json.

🚀 Getting started (quick‑start)

# 1️⃣ Install on your platform (example for Claude Code)
/plugin marketplace add Egonex-AI/Understand-Anything
/plugin install understand-anything

# 2️⃣ Analyze a project (first run may use many LLM tokens)
/understand            # creates .ua/knowledge-graph.json

# 3️⃣ Open the interactive dashboard
/understand-dashboard

For other platforms use the universal installer:

curl -fsSL https://raw.githubusercontent.com/Egonex-AI/Understand-Anything/main/install.sh | bash -s <platform>

Supported <platform> values include codex, gemini, vscode, kiro, cursor, etc.

💡 Typical use cases

On‑boarding – New hires run /understand and instantly explore the whole codebase visually.
PR reviews – Commit the generated .ua/*.json graph; reviewers can see impact analysis without re‑running the pipeline.
Business‑logic discovery – /understand-domain extracts high‑level processes (e.g., payment flow) from the code.
Documentation of wikis – Point /understand-knowledge at a markdown wiki to get a concept graph for internal knowledge bases.
Local, privacy‑first analysis – Plug a local LLM (e.g., Ollama) via the --model config to avoid sending source to the cloud.

📦 Sharing the graph

The graph is plain JSON, so you can commit it (excluding intermediate/ and diff‑overlay.json). Large graphs can be stored with git‑LFS. Anyone can view the dashboard without an LLM by running:

npx https://github.com/Egonex-AI/Understand-Anything/releases/latest/download/understand-anything-viewer.tgz /path/to/project

This serves the graph locally on http://127.0.0.1:5173.

🤝 Contributing & community

Run tests with pnpm --filter @understand-anything/core test.
Follow the usual fork → branch → PR workflow.
Community videos are welcomed; open an issue or PR to be featured.

📜 License

Bottom line: Understand Anything lets you turn any sizable codebase (or markdown knowledge base) into a searchable, visual knowledge graph powered by a hybrid static‑analysis + LLM pipeline, with support for many AI coding assistants and a privacy‑friendly local‑only mode.