rtk

rtk-ai/rtk

CLI proxy that reduces LLM token consumption by 60-90%

Why pick itSingle Rust binary, zero dependencies

Install

GitHub: rtk-ai/rtk
Website: rtk-ai.app
Stars: ★ 84.5k
License: Apache-2.0
Language: Rust
Last push: Oct 23, 2026
Category: Run locally

Good for

Filter and compress command output before it reaches the LLM context to cut token use
Automatically rewrite agent commands like git status to their rtk versions
Work with Claude Code, Copilot, Gemini CLI, Codex, Cursor and many other agents

Getting started

Run brew install rtk on macOS/Linux, winget install rtk-ai.rtk on Windows, or use the curl install script
Run rtk gain and confirm it shows the savings dashboard (another project named rtk exists on crates.io)
Run rtk init -g (with a flag such as --codex for other agents) and restart your AI tool

Before you install

The hook only fires on Bash tool calls; built-in tools like Read, Grep and Glob bypass it. Production use is free, but offering RTK to third parties as a commercial product or service needs a separate licence.

Summarized by AI from the project docs; check the official docs before relying on it.

Similar tools

unslothRun locally · ★ 78.5k
GUI app to run and train large language models and diffusion models locally
OllamaRun locally · ★ 185k
Download and run open-weight models like Qwen, DeepSeek, Gemma and gpt-oss locally with one command, with an OpenAI-compatible API.
Open WebUIRun locally · ★ 157k
A self-hosted ChatGPT-style web UI that connects to Ollama and any OpenAI-compatible API, with built-in RAG, web search, multi-user support and MCP.
Cherry StudioRun locally · ★ 52.9k
A cross-platform desktop AI client that puts many cloud models and local Ollama in one interface, with a built-in knowledge base, MCP, agents and 300+ assistant templates.
JanRun locally · ★ 45.1k
A ChatGPT-style desktop app that runs fully offline, with built-in model downloads and the llama.cpp engine; can also switch to cloud APIs.