GitHub radar
Cut your AI agent's terminal token usage by up to 90%
This is a tool that trims long terminal output before your AI agent reads it. Install it and your agent burns fewer tokens on logs and errors, keeping more context.
It's a small command-line proxy: it catches the output of git, tests, and builds, and shortens it to what matters.
When it helps
Useful if you rely on an AI agent (Claude Code, Cursor, etc.) and watch long test logs or git diffs eat your context window. Skip it for a small project with short output.
Pros
- Cuts command output by up to 90%, saving tokens and context.
- Covers 100+ commands: git, test runners, Docker, AWS CLI.
- Hooks into Claude Code, Cursor, and other popular AI agents.
- No GPU needed — runs on an ordinary laptop.
Cons
- Install it first, then set up a hook for your agent.
- Only compresses terminal commands, not every action your agent takes.
How to set it up — step by step
- 1Install RTK: on Mac and Linux run `brew install rtk`; on Windows run `winget install rtk-ai.rtk`.
- 2Check the install: `rtk --version` should print a version, and `rtk gain` should open a savings dashboard.
- 3Open your AI agent (Claude Code, Cursor, etc.) in your project folder and set up the hook from the README's integration section.
- 4Send your agent the text from the block below — it will configure RTK and show you how to check it works.
- 5Restart your AI agent so the hook takes effect.
- 6Run any command, like `git status`, and confirm the output is shorter now.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install RTK from https://github.com/rtk-ai/rtk — follow the README (brew install rtk or curl script), then configure the Claude Code hook per the README integration section. Show me how to verify RTK is working (`rtk gain`).
Runs on a regular laptop — no GPU needed.
Open on GitHub▌ More finds