Explore Project Relationships with Understand Anything: An LLM‑Powered Code Knowledge Graph

When developers join a new team, the sheer size of a codebase—hundreds of files and hundreds of thousands of lines—makes understanding the system feel like an archaeological dig. Traditional documentation is often outdated, and reading code line‑by‑line is inefficient.

Understand Anything addresses this problem by leveraging Claude Code’s multi‑agent architecture. It fuses a large language model (LLM) with static code analysis to automatically extract files, functions, classes, and dependencies, then builds an interactive knowledge graph that can be explored, searched, and queried in natural language.

Installation steps: npm install -g @anthropic-ai/claude-code After installing Claude Code, the author switches the model to MiniMax 2.7 via the CC Switch UI, then adds the Understand Anything plugin:

/plugin marketplace add Lum1104/Understand-Anything /plugin install understand-anything

The demo uses the open‑source CRMEB Java e‑commerce project (≈1,000 Java files). After cloning the repository, the author opens it in IDEA and runs the command: /understand The first phase scans the project, reporting file count and complexity. In the second phase, five parallel agents create 43 batches to process the files, generating temporary files and finally producing .understand-anything/knowledge-graph.json. Screenshots show the batch progress and the final JSON file.

To visualise the result, the author runs: /understand-dashboard This opens a web dashboard where the codebase is displayed as a colour‑coded hierarchical graph. Nodes can be clicked to view source code, dependencies, and concise natural‑language explanations.

Key features highlighted include semantic search (e.g., “how is authentication implemented?”), change‑impact analysis before committing, and guided learning paths that adapt the UI for developers, product managers, or designers. The tool also provides commands such as /understand-diff, /understand-explain <file>, and /understand-onboard to generate onboarding guides.

The author notes that very large repositories may strain the current setup and that the MiniMax 2.7 model sometimes lacks the depth needed for complex analysis; a stronger model like OPUS 4.6 would improve results.

Overall, Understand Anything transforms a tangled code maze into an explorable knowledge map, making system comprehension faster for junior developers, product managers, designers, and AI‑assisted development workflows.