Map your codebase to help new teams understand and share context
If you have ever faced a 200,000-line codebase and not known where to start reading, this tool offers a solution. It converts all files, functions, and dependencies in your code into an interactive knowledge graph that you can explore visually and query. Beyond simply showing structure, it connects business logic and architecture layers, reducing the time needed to understand how the entire codebase fits together.
Explore code structure like a map
The tool creates nodes for every file, function, and class in your project and displays them on an interactive dashboard. Clicking a node reveals a plain-language explanation of the code, its relationships, and a guided tour with a defined learning order. Architecture layers (such as API, service, and data) are color-coded to make the system hierarchy intuitive. You can also use semantic search to quickly find code snippets responsible for specific features, such as asking, 'Where is the authentication handling?'
Connect business logic and knowledge bases
Beyond code structure, the tool provides a domain view that shows how the code maps to actual business processes. It arranges domains, flows, and steps in a horizontal graph to visualize the mapping between code and business logic. When analyzing a Karpathi-pattern LLM wiki, it extracts wikilinks and categories, discovers implicit relationships and entities, and converts them into a graph of interconnected ideas. This process turns a documented knowledge base into an explorable knowledge graph.
Works across various AI coding platforms
You can use the tool with various AI coding platforms, including Claude Code, Codex, Cursor, Copilot, and Gemini CLI. Claude Code installs it via the plugin marketplace, while other platforms detect it automatically after running an installation script or cloning the repository. It supports platform-specific invocation methods (such as slash commands) and allows you to specify local models (like Ollama) for use in privacy-focused or enterprise environments. After installation, you start the analysis with the /understand command, which supports incremental updates that reanalyze only changed files by default.
Team sharing and dashboard usage
The generated graph is saved as a JSON file, making it easy to share among team members. If you commit the .ua/ directory, other team members can view the dashboard without running the pipeline, which is useful for onboarding or PR reviews. You can view the dashboard locally with just Node.js, even without Claude Code or an LLM. All data is provided read-only from the local disk and is not sent externally. You can also set up a hook to automatically update the graph on every commit, maintaining consistency between the code and the graph.
Things to check before use
The initial analysis scans the entire codebase, which can result in significant token consumption for large projects. Considering this, it is recommended to use a token subscription plan or a local model. Structural analysis is handled deterministically using Tree-sitter, so the same code always produces the same result, while semantic analysis is handled by an LLM. The project is released under the MIT license, and the local viewer runs in environments with Node.js 18 or higher. For complex monorepos, it provides an option to narrow the analysis scope to specific subdirectories.