Zach Ahn · 2026-08-19 · notable
Vomit — a local model rewrites Claude Code's replies before you read them
Vomit is a Go tool that hooks into Claude Code, buffers each reply and hands it to a local model to be rewritten in plainer language before it reaches the screen. Zach Ahn runs gpt-oss:20b for the rewriting pass.
Vomit pipes Claude Code's output through a small local model so the prose comes back plain.
What is it?
Vomit hooks into Claude Code and rewrites what the model says before you see it. Zach Ahn built it after deciding that Claude 5's default style — dense jargon and stretched metaphors — got in the way of the work, and describes the result as an art project that is new and lightly tested. The tool is written in Go and licensed GPL-3.0.
How does it work?
The tool buffers Claude Code's output, forwards each message to a local LLM, waits for the rewrite, then renders the new version through the MessageDisplay hook. Ahn uses gpt-oss:20b for that pass, so the rewriting stays on the machine. He notes some of the original phrasing still slips through, including plural pronouns like 'we'.
Why does it matter?
Complaints about a model's writing style usually get answered with a longer system prompt; Vomit answers with a second model instead. The Hacker News thread reached 154 points and 271 comments, most of them arguing about whether the style is a training side effect of agentic coding work or a dialect you eventually learn to read.
Who is it for?
Claude Code users
Try it
https://github.com/zachahn/vomit