Skip to content

Introduction

tok0 is a high-performance CLI proxy that compresses verbose shell command output before it reaches your LLM. Typical savings are 60–95% on git, cargo, npm, docker, kubectl, and ~250 other commands.

It’s a single Rust binary (≤8 MB stripped), single-threaded by design, and adds less than 2 ms of overhead per command.

Modern coding agents pipe tool output straight into the model context. A single npm install can burn 8,000 tokens on download bars; a cargo build can dump 30,000 tokens of compiler noise. tok0 sits between the agent and the command and emits the structurally important part — errors, summary counts, final status — while preserving exit codes and behaviour.