February 22, 2026
Folders are canceled
Browse Code by Meaning
A new tool lets you navigate projects by vibes; coders hype and hurl skepticism
TLDR: A new “semantic navigator” lets you explore code by meaning instead of folders. Comments split between hype for non-chat tools and worries about cost, speed, and sending code to an AI, turning it into a debate over interfaces vs chatbots.
The dev behind Grace just dropped a “semantic navigator” that lets you browse any repo by meaning, not folders, and the comments went nuclear. Fans cheered the move outside the chat box, saying chatbots drown you in prose and delays. The demo shows auto-labeled clusters, summaries for small projects, and nested groups for big ones—up to around 10,000 files. The big reveal: it uses OpenAI under the hood, defaulting to gpt-5-mini for cheaper labeling, with a faster model available at a 7× price tag. Cue the community chorus calling it “Google Maps for your repo,” and linking the GitHub repo like it’s hot.
Then the drama kicked in. Skeptics side-eyed the OPENAI_API_KEY requirement: who’s comfortable sending company code to an AI, and who’s footing that bill? One camp argued folders exist for a reason—“just name things well.” Another fired back that directory trees break under scale, and a meaning-first view is the future. Meme-lords declared “directories are canceled,” while cynics joked, “Find meaning in my spaghetti code? Good luck.” Some asked whether “clusters” are trustworthy or just vibes that hallucinate. The thread devolved into a spicy showdown: better interfaces vs chat agents, speed vs cost, privacy vs convenience—plus a splash of “Marie Kondo for code” energy. Peak internet.
Key Points
- •An open-source semantic navigator lets users browse repositories by meaning, not directory hierarchy.
- •The tool uses OpenAI models (default: gpt-5-mini; optional: gpt-5.2) to label semantic clusters.
- •Small repos (≤20 files) are summarized per-file; medium and large repos show top-level and nested clusters.
- •It can process approximately 10,000 files within minutes on a modern MacBook.
- •Implementation involves embedding files, recursive clustering, labeling nodes, and rendering a tree view; it works on any text documents.