The Free Tool That Turns 200,000 Lines of Code Into a Map Your Whole Team Can Read

Understand-Anything is an MIT-licensed plugin with 14,700 GitHub stars that turns any codebase into an interactive knowledge graph. It replaces Sourcegraph Cody Enterprise, which moved to an enterprise-only model at $59 per user per month in 2025, and addresses the $20,000 to $80,000 cost companies absorb every time a developer needs weeks or months to get productive in a new codebase.

Understand-Anything, a free MIT-licensed plugin sitting at 14,700 GitHub stars, turns any codebase into an interactive knowledge graph that any team member can explore, search, and ask questions about directly inside their AI coding tool of choice. It addresses the same problem Sourcegraph Cody Enterprise charges $59 per user per month to solve: the time and money lost when someone new, or even an experienced developer moving to an unfamiliar part of the system, cannot see how a large codebase fits together. Sourcegraph ended its free and Pro tiers in July 2025. A 20-person engineering team that previously used Cody Pro now faces $1,180 per month to keep that code intelligence capability. Understand-Anything runs free.

What it actually does

When you install the plugin and run /understand, a multi-agent pipeline analyzes your project. Six specialized agents scan the files, extract every function, class, import, and dependency, identify architectural layers, build guided learning tours, and validate the resulting graph for completeness. The output is a knowledge graph stored as JSON in your project folder, which then renders as an interactive visual dashboard in your browser.

The dashboard is not a static diagram. You can pan and zoom across the graph, click any node to see a plain-language explanation of what it does and what it connects to, run fuzzy and semantic searches across the entire codebase, and ask questions in natural language. "How does the payment flow work?" returns a guided path through the relevant nodes rather than a list of files to open.

A diff impact view lets you run /understand-diff before committing to see which parts of the system your edit touches, including connections you might not have anticipated.

The real cost it displaces

The Sourcegraph comparison is the obvious framing, but the more significant number is what codebase opacity costs in developer time. Research on developer onboarding consistently puts the total cost of getting a new engineer productive at $20,000 to $80,000 across the first 90 days, with a meaningful portion of that coming from the senior engineers pulled away from their own work to explain architecture. A typical onboarding scenario involves 15 to 20 hours per week of senior engineer mentoring time over 8 to 12 weeks. At average senior engineer rates, that is $8,000 to $18,000 of capacity that disappears into explanations of how things connect.

A committed knowledge graph in the repository does not replace that mentoring entirely, but it handles the low-level orientation work: what files exist, what they depend on, what the architectural layers are, which flows handle which business domains. That leaves mentoring time for the decisions and judgment calls that actually require a human.

For teams that hire external consultants to document undocumented codebases, the displacement is more direct still. A two-week architecture review costs $15,000 to $40,000 and produces a document that is out of date within months. A committed knowledge graph updates with each pull request.

How it plugs in

Understand-Anything works as a native plugin for Claude Code, and also installs into Cursor, GitHub Copilot in VS Code, Codex, Gemini CLI, and a handful of other AI coding platforms. Installation via Claude Code is two commands: add the plugin from the marketplace and install it. For other platforms, a one-line shell command handles the setup.

The graph is committed to the repository alongside the code. Once the initial graph is generated, teammates skip the analysis pipeline entirely and open the dashboard directly against the committed JSON. Incremental updates run automatically after each commit through a post-commit hook, so the graph stays current rather than becoming a snapshot that drifts.

The JSON format means the graph is just another file in version control, diff-able, searchable, and present in every branch. A code reviewer can open the graph against a feature branch and immediately see how the proposed changes sit within the broader system.

What the setup actually requires

There is no server to run and no infrastructure to maintain. The plugin runs the analysis pipeline locally using whichever AI coding tool you already have. The plugin itself is free, but it delegates the analysis work to the AI model behind your tool, so you are drawing from your existing token budget. Teams already paying for Claude Code or Copilot are adding capability without adding cost. Teams that have not adopted an AI coding tool yet would need one.

The analysis time scales with codebase size. A small project finishes in a few minutes. A 200,000-line monorepo takes longer depending on the provider's rate limits. The pipeline runs up to five file analyzers in parallel and processes 20 to 30 files per batch, which keeps large codebases manageable.

Where it falls short

The knowledge graph quality depends entirely on the AI model doing the analysis. Files that are well-structured, well-named, and follow conventional patterns produce rich, accurate nodes with useful descriptions. Older codebases with inconsistent naming conventions, large functions that do multiple things, or significant amounts of generated or minified code produce thinner analysis. The graph reflects how readable the code is.

The interactive dashboard requires opening a local web server. There is no hosted version with sharing links or team access controls. If a product manager or engineering manager wants to explore the graph, they need the repository locally and the plugin installed. For organizations where non-engineers regularly review system architecture, this friction is real.

The business domain view is powerful but only as good as the domain boundaries the code itself expresses. A codebase that was never designed with clear domain separation produces a graph that accurately reflects the tangle rather than clarifying it. Enterprise features around role-based access and centralized hosting are not present, which matters for companies with compliance requirements around who can see what in the codebase.

14,700 stars for something that used to cost a lot

There is a pattern in open-source tools that become viral on GitHub: they tend to replace something that a large, well-funded company made expensive right before the open-source version got good enough to matter. Sourcegraph spent years making Cody into a sophisticated product, then moved it to enterprise-only pricing in 2025. The month after that decision, codebase intelligence tools started showing up in the GitHub trending lists.

Understand-Anything earned 4,697 stars today. That is a specific kind of signal: not people bookmarking something interesting, but people installing something immediately useful. The knowledge graph is committed to the repository by default, which means anyone who uses it once starts giving their teammates the same capability without requiring anyone else to install anything. That is how genuinely useful tools spread inside organizations. Someone onboards a new hire faster, the new hire tells someone else how they figured out the codebase so quickly, and suddenly a free JSON file in the repository is doing work that used to belong to a $59-per-seat subscription.