Launched this week
Understand Anything
Onboard to any codebase in minutes, not weeks.
16 followers
Onboard to any codebase in minutes, not weeks.
16 followers
Understand Anything is an open-source project with 15k+ GitHub stars. It turns any repo into an interactive knowledge graph you can navigate, search, and learn from. It pairs tree-sitter static analysis as ground truth with LLM-generated explanations, so you get real code structure plus readable architecture narratives. Explore codebase maps, guided tours, file-level walkthroughs, and inline code reading. Works with Claude Code, Cursor, and opencode via slash commands. No vendor lock-in.
An excellent project! I tried using it to understand my team's codebase, and it significantly improved my efficiency. I hope more features will be added in the future, such as:
It struggles with very large codebases. My codebase has tens of thousands of files, generating a knowledge graph of dozens of megabytes. I hope sharding support can be added.
/understand-domain relies too much on the agent's summarization ability, resulting in insufficient detail about business processes, especially for the large codebase I mentioned above.
I'm planning to provide this tool to product managers as well, so they can understand the current business state of our long-standing codebase. However, many product details cannot be extracted. I hope you can consider ways to generate more business domain details from the codebase for product managers to use.
@tylor_liu Got it!
For 1, you can provide a folder to the /understand command
For 23, we will look into how to provide a more detail domain knowledge!
If you've ever tried reproducing a research paper, you know the pain: clone the repo, open 40 Python files, and still have no idea where the core model or loss function actually lives.
I tried Understand Anything on a multimodal medical imaging repo (~20k LOC), and the graph finally made the architecture click for me.
At first it looked dense, but once I started exploring, it became insanely useful. Every file has a plain-English summary that's clearer than most docstrings, and tracing the data flow from dataloader → preprocessing → model took minutes instead of hours.
Honestly feels much closer to “understanding” a codebase than just searching through it.
@jamie_hm Yeah, code is everywhere, and will be powerful if we can understand them!