Most AI coding tools operate in silos. Claude Code has its own context. Copilot has another. Cursor, Codex, Gemini CLI—each carries a separate understanding of your codebase.
Until this week, that meant switching tools meant losing context. Not anymore—at least in theory.
The new open-source project Understand-Anything claims to turn any code into a single interactive knowledge graph. Explore, search, ask questions. And it works across Claude Code, Cursor, Copilot, Codex, and Gemini CLI.
That is the promise: one graph to rule them all. A unified, queryable map of your codebase accessible from any assistant.
The promise of a shared code brain
Imagine asking Claude Code, "What calls this deprecated function?" and getting an answer that also highlights the same dependency in Copilot when you switch tools. No re-indexing. No lost context.
Or using Gemini CLI to ask plain-English questions about a gnarly algorithm, with direct links to the relevant code nodes. Then plugging into Cursor to visually navigate the call hierarchy.
A team might integrate it with Copilot in VS Code to visually trace class hierarchies. A new developer could search for all instances of an API endpoint, seeing a map of usage across the codebase via Codex integration.
The core proposition is deceptively simple: an interactive model of your code that any AI assistant can tap into. It's not just another visualisation tool. It's an attempt to solve context fragmentation.
What the project actually claims
The GitHub repository is refreshingly straightforward. Its entire description reads:
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.
That's it. No architectural diagrams. No deep dives into graph generation. No language compatibility matrix.
We don't know whether the graph is built via static analysis, LLM parsing, or some hybrid approach. Performance on large monorepos remains a question mark. And "and more" hints at ambition without specifying integration depth.
But the bullet points are enough to make the intent clear: a universal abstraction layer for code understanding, consumed through whichever AI assistant you prefer.
Why context fragmentation hurts
Multi-tool workflows are now the norm. You might debug in Copilot, refactor in Cursor, and generate docstrings with Claude Code. Each shift costs you the mental model you'd built in the previous tool. The assistant forgets what the other assistant knew.
A shared knowledge graph could bridge that gap. It wouldn't magically align model reasoning, but it would give each tool the same structural map of the codebase. That's a meaningful improvement over the current state, where each tool independently reconstructs its own version of your code.
The project touches a real pain point. Even if today's implementation is thin, the concept is worth watching.
Holding the scepticism
Early-stage projects deserve enthusiasm tempered with honesty. Understand-Anything currently offers a vision more than a verified solution. No examples of actual graph generation sit in the repo. No queries or visualisations demonstrate the interactive experience. Community adoption isn't measurable yet.
But this isn't unusual for projects that are just surfacing. The interesting bit isn't what the codebase does right now. It's the problem statement it pins to the wall: context switching across assistants is a tax we should stop paying.
Which chore in your multi-tool workflow would you most want unified by a knowledge graph?
Top comments (0)