rtk

Overview

rtk is a CLI proxy that sits between your terminal and your AI coding tool. When you run a command inside Claude Code or Cursor, rtk intercepts the output and compresses it — removing redundancy, collapsing verbose stack traces, trimming repeated patterns — before the model processes it. You see the full output. The model sees the compressed version.

The compression is transparent in both directions. Claude never knows it happened. You never have to think about it. You run the same commands you always ran; rtk handles the token reduction automatically.

At 32K+ GitHub stars and 813 new yesterday, rtk has become a standard efficiency layer in high-volume Claude Code setups. The primary use case is developers who run many AI-assisted iterations per day and have started noticing the token cost of verbose command output.

Key Capabilities

Transparent interception: rtk rewrites commands at the shell level before execution. git status becomes rtk git status — but only in the AI’s view. You never see the rewrite. This approach avoids modifying any AI tool’s configuration or API surface.

Compression that preserves signal: The compression logic is designed to preserve actionable information — error messages, file paths, changed lines — while collapsing noise. A 155-line cargo test run becomes 3 lines (98% reduction) without losing the actual failure message.

Broad tool support: Claude Code, Cursor, GitHub Copilot, Gemini CLI, Codeium, Aider, Continue, Tabnine, OpenClaw, and three others. The proxy intercepts at the shell level, so any tool that runs terminal commands benefits.

Zero-dependency binary: rtk ships as a single Rust binary with no runtime dependencies. It does not require Node, Python, or any package manager beyond the initial install.

Installation

# Install
brew install rtk-ai/tap/rtk   # macOS
# or: cargo install rtk

# One-time global setup
rtk init -g

# Restart Claude Code (or your IDE) — done

After rtk init -g, every AI coding session automatically routes through the proxy. No per-project configuration is needed.

Performance Numbers

From the rtk README (two weeks of daily use):

Command	Before	After	Reduction
git status	119 chars	28 chars	76%
cargo test	155 lines	3 lines	98%
Total over 2 weeks	~90M tokens	~10M tokens	89%

Actual savings vary by project type. Projects with large dependency trees and verbose test output (Rust, Java, Go monorepos) see the largest gains. JavaScript projects with short command output see smaller but still meaningful reductions.

Who It’s For

Developers running heavy Claude Code or Cursor workflows who have noticed token consumption growing with project complexity. The break-even is low — if you run more than 50 AI-assisted commands per day, rtk typically pays for itself in reduced API costs within the first session.

Also useful for teams running Claude Code on shared API keys where token budgets are actively managed. rtk’s compression is deterministic, so results are reproducible across team members.