Vomit

// Overview

Vomit is a small Go command-line tool that sits between Claude Code and your eyes. It hooks into Claude Code and rewrites the agent's output inside the Claude Code terminal UI, handing each reply to a local LLM that restates it in plainer English. It is fully local with no telemetry and no external dependencies — the rewriting model runs on your own machine.

The design has an honest limit the README states up front: the local model only sees what Claude tries to communicate, not the actions it took or the files it touched, so the rewritten text can hallucinate a little. The original output is never lost — pressing Ctrl-O inside Claude Code shows Claude's own text. The author also flags the project as vibe-coded, tested only on macOS, and fairly slow, which is the trade for running the rewrite pass locally.

Any local server that speaks the OpenAI API should work as the rewriting backend; the README names Llama.app, Ollama and Apfel, and suggests GPT-OSS 20B through Llama.app if you do not already have a local model set up. Beyond the hook-based mode there is an older non-invasive mode, now deprecated, that tails a session and translates it alongside the real UI. Vomit is licensed GPL-3.0.

// What it does

• Hooks into Claude Code and replaces its output in-place inside the terminal UI
• Runs the rewrite pass on a local LLM — fully local, no telemetry, no external dependencies
• Works with any OpenAI-API-compatible local server, including Llama.app, Ollama and Apfel
• Ctrl-O inside Claude Code brings back Claude's original, unrewritten output
• A deprecated side-by-side mode (vomit list / vomit tail) that follows a session without replacing its output
• Single Go binary installable with go install

// Getting started

You need a local LLM server running first — the README recommends Llama.app with GPT-OSS 20B if you have nothing set up. Then install the binary, point it at your server, and wire it into Claude Code's hooks.

go install github.com/zachahn/vomit@latest

vomit init

vomit scrub -claude

vomit list
vomit tail
vomit help

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Cut down the volume of text a coding agent puts on screen during a long session
• Read an agent's narration in plainer language while keeping the original one keystroke away
• Put a local model to work on a small, low-stakes job that never leaves your machine
• See a minimal worked example of Claude Code's hook system rewriting output in the terminal UI

// How Vomit compares

Vomit alongside other open-source agent skills & plugins tools AI/TLDR tracks, ranked by GitHub stars.