How a Single Command Cuts AI Coding Token Usage from 210K to 23K

Token waste in AI‑assisted coding

During a typical 2‑hour AI coding session the CLI output alone generates about 210,000 noisy tokens . For a 10‑person team this translates to roughly $1,750 per month spent on tokens that never contribute to the model’s reasoning.

RTK (Rust Token Killer)

RTK is an open‑source CLI proxy written in Rust. It sits between an AI tool and the terminal, intercepts commands such as git status or cargo test, parses the multi‑line output, extracts a concise summary (e.g., “2 test cases failed”), forwards only that summary to the model, and saves the full log locally for later inspection.

Measured token savings

git status: 3,000 → 600 tokens (‑80%)

cargo test: 25,000 → 2,500 tokens (‑90%)

pytest: 8,000 → 800 tokens (‑90%)

ls / tree: 2,000 → 400 tokens (‑80%)

In a full 2‑hour session the total token count drops from 210,000 to about 23,000 , an ~80% reduction.

Installation

# macOS (recommended)
brew install rtk

# Linux
curl -fsSL https://raw.githubusercontent.com/rtk-ai/rtk/refs/heads/master/install.sh | sh

Initialization

rtk init -g

The command automatically detects the AI assistant in use and configures the proxy.

Usage

Run commands as usual; RTK runs in the background. For example, git status is transformed into rtk git status, and the AI receives only the filtered output.

Supported tools

RTK works with common development commands (git, cargo, pytest, ls/tree, etc.) and integrates with various AI assistants.

Conclusion

Feeding the model only the essential information reduces token consumption and can improve reasoning accuracy. RTK translates human‑oriented terminal output into the minimal data the AI actually needs.

GitHub shows over 39,000 ★ for the project, which is released under the MIT license.

Project URL: https://github.com/rtk-ai/rtk