rtk: the Rust proxy that shrinks your coding agent's command output

rtk 1.0

Your coding agent does not burn most of its tokens on thinking. It burns them on reading command output: long git status lists, full test logs, and verbose diffs. rtk, short for Rust Token Killer, sits between your agent and the shell and shrinks that output before the model ever sees it. A 40 line git status arrives as three lines, and test output shows failures only. It is a single Rust binary with no dependencies, covering more than 100 commands with under 25 milliseconds of overhead, and it installs with one rtk init command for 18 tools including Claude Code, Cursor, Codex, and Gemini CLI. Version 1.2.0 shipped on 8 November 2026, and the repo now has 86,534 stars.

What you get

• rtk gain shows exactly what you saved, per command and per day, so the benefit is measured on your machine, not promised in a benchmark
• rtk bypass runs any command unfiltered when you need the full output back
• It never blocks: if a filter fails you get the raw output, and the command exit code passes through unchanged
• Installs with Homebrew, winget, or one install script, then one rtk init plugs it into your agent

Start with the defaults, check rtk gain after a week, and only then decide whether to go further with pipelines.

Repo: https://github.com/rtk-ai/rtk

Quick poll

What would you use rtk 1.0 for first?

• Trimming test output
• Shrinking git status and diffs
• Measuring the savings with rtk gain
• Watching it for now

Vote in the comments.

---

If you found rtk 1.0 interesting, I write about tools like this every week in The AI-Native Engineering Leader, what actually works for engineering leaders building with AI agents, and what to skip.