Vomit: Claude 5 Output Cleaner
Anyone who uses Claude Code knows the pain: Claude 5's output is often bloated, tacking hundreds of lines of explanation onto a simple function edit. zachahn built Vomit, a Go CLI tool that uses a lightweight LLM to clean up Claude 5's verbose output and save your tokens.
Install it, run one command, and it automatically strips out the filler text from Claude's responses, leaving only the useful code and key notes. One HN commenter called it "exactly what Claude Code users need." The 193 upvotes say a lot about how frustrated people are with Claude 5's wordiness.
GitHub: https://github.com/zachahn/vomit
ParqDB: Vector Search Right in Your Browser
ParqDB does something genuinely cool: it lets you run vector search on Parquet files directly in the browser. No database setup, no backend server needed. Feed Parquet files to the browser over HTTP, and it uses MiniLM embeddings plus WebAssembly to compute vectors locally, with HTTP Range requests for efficient retrieval.
Data never leaves your browser. For data engineers who work with Parquet regularly, this is a lightweight way to explore large datasets without spinning up a vector database.
URL: https://search.parqdb.io/
DumpSort: Auto-Rename and Zip Your Screenshots
The worst part about accumulating screenshots? File names like "Screenshot 2026-08-21 at 14.32.05.png" that tell you nothing. DumpSort is a browser-only tool: drop in your screenshots and PDFs, and it automatically recognizes the content to generate meaningful file names, then packages everything into a zip for download.
Everything happens in the browser — no files get uploaded anywhere. For anyone who takes screenshots for notes or document organization, this little tool saves a surprising amount of manual renaming time.