vomit
Clean up Claude's token vomit with a separate LLM. Save your tokens, Opus is hopeless
What is it?
What it is
A Go-based CLI tool that intercepts Claude's raw token output and translates it into readable English using a locally hosted LLM.
Why it exists
To provide a privacy‑preserving, offline way to understand Claude's verbose token stream without external telemetry or dependencies.
Who should use it
Claude Code users who find the default output hard to read and want a local translation layerDevelopers running Claude Code sessions on macOS who already have a local LLM (Llama.app, Ollama, or any OpenAI-API-compatible server) set upUsers with privacy concerns who need a fully local, no-telemetry tool for processing Claude's session outputGo developers comfortable installing via `go install` and configuring CLI tools via a setup subcommand
Who should avoid it
Users who do not have a local LLM installed and configured (a hard prerequisite for the tool to function)Non-macOS users (the project has only been tested on Mac and may not work on Linux or Windows)Users who require reliable, production-grade translation of Claude's output — the README explicitly warns the LLM hallucinates, is slow, and may miss messages entirelyAnyone unwilling to lose information, since Vomit writes nothing during runtime besides TMPDIR files and recommends pairing it with a separate viewer (AgentsView) to recover the original outputUsers uncomfortable with vibe-coded, single-author, hobbyist tools with minimal testing
Stars over time
No change stars in the last 1 day
How it works
A quick walkthrough in plain English
How vomit works in 4 steps
Step 1 of 4
Something triggers the flow
A schedule, webhook, or manual click tells vomit to start.
Features
Advantages
- Open source under GPL-3.0
- No external dependencies
- No telemetry
- Can use various local LLMs
- Allows listing and following sessions
- Simple CLI interface
Disadvantages
- Slow performance
- Only tested on macOS
- May miss Claude's messages
- Requires local LLM setup
- Potential hallucinations
- Writes files to TMPDIR
Installation
native
go install github.com/zachahn/vomit@latest; vomit init; vomit scrub -claude
FAQ
How do I install Vomit?
Run `go install github.com/zachahn/vomit@latest` to install the binary to your GOPATH, then run `vomit init` to configure your local LLM connection.
What does Vomit do with Claude's output?
Vomit intercepts Claude's token stream (the 'token vomit') and pipes it through a locally‑run LLM to translate the tokens into readable English, without accessing any of Claude's actions or files.
How can I view or follow a Claude session with Vomit?
Use `vomit list` to see available session identifiers, then `vomit tail [<session_identifier>]` to translate tokens for a specific session or follow the latest one.
Which local LLMs are compatible with Vomit?
Vomit works with any LLM that exposes an OpenAI‑compatible API, such as models served by Llama.app or Ollama. The README recommends using Llama.app with the GPT‑OSS 20B model.
Is Vomit safe and private?
Yes. Vomit runs fully locally, has no telemetry, and only sees the tokens Claude tries to communicate; it does not access your files, actions, or external services, though the local LLM may hallucinate.
Featured in Videos
YouTube tutorials and walkthroughs for vomit
Alternatives
Similar projects ranked by category, topics, and text overlap.
