RTK: A High-Performance Rust CLI Proxy That Slashes AI Agent Token Costs The rtk-ai project released RTK (Rust Token Killer), an open-source CLI proxy written in Rust that intercepts shell commands and compresses terminal output before AI coding agents ingest it. The tool applies tailored filters for over 100 developer tools, including git, pytest, cargo, and docker, and claims to cut bash output token consumption by 60% to 90% with sub-10ms overhead. It integrates with agents such as Claude Code, Cursor, and Codex via lifecycle hooks and runs fully offline. As developers increasingly delegate terminal tasks, testing, and debugging to AI coding agents such as Claude Code, Cursor, Codex, and OpenHands , context window saturation has become a major engineering bottleneck. Standard developer tools—such as git diff , test runners, and package managers—were designed for human terminal readability, often generating verbose logs filled with progress indicators, decorative ASCII headers, and repetitive success statuses. When fed directly into an LLM's context window, these verbose logs rapidly consume input token quotas, degrade prompt caching efficiency, and crowd out essential architectural context. RTK Rust Token Killer is an open-source, high-performance CLI proxy developed by rtk-ai. Written in Rust with sub-10ms overhead, rtk intercepts shell commands and compresses terminal stdout before your agent ingests it. rtk functions as a transparent proxy between your coding agent and your system shell: Agent ➔ Shell ➔ Git/Cargo ➔ 1,000 lines of raw output ➔ Agent context Agent ➔ Shell ➔ RTK Filter ➔ 25 lines of structured, actionable context ➔ Agent context By stripping noisy boilerplate, grouping recurring errors, and collapsing passing test suites into concise summaries, rtk reduces bash output token consumption by 60% to 90% across standard developer workflows. Engineered for raw speed, rtk compiles to a single native binary. It introduces virtually no measurable latency to developer command pipelines while operating completely offline without external network calls. rtk applies specialized reduction logic tailored to over 100 developer tools: Out of the box, rtk includes tailored filters for: git status , one-line commit logs, and header-stripped git diff . pytest , cargo test , vitest , jest , go test , and playwright collapsing passing suites and highlighting failures only . tsc , ruff , eslint , cargo clippy , and sqlfluff . docker ps retaining only critical status fields. RTK integrates seamlessly into AI agent environments using native lifecycle hooks: Install via Homebrew macOS / Linux brew install rtk Or via Windows Package Manager winget install rtk-ai.rtk Initialize hook for your AI agent rtk init -g Claude Code / Copilot rtk init -g --agent cursor Cursor rtk init -g --codex Codex Once initialized, standard agent bash executions automatically route through rtk , keeping your prompt context lean and responsive. Context window space is the most valuable real estate in autonomous coding workflows. By filtering terminal noise at the source, RTK prevents context pollution, accelerates agent response times, and cuts operational API expenses. Ready to streamline your agent's terminal context? Check out the RTK GitHub Repository https://github.com/rtk-ai/rtk .