Dev Built a Stick Shift to Drive Claude Code
Typing terminal commands is apparently too mundane. When an engineer saw a proprietary gadget turning AI prompt switches into manual gear shifts, they did the only sensible developer move: built an open-source transmission that revs straight inside the command line.
Frustrated by a closed-source, paid macOS utility called Model Shift, developer Ilya Tuzov decided that executing standard slash commands to swap language models lacked tactile drama. Rather than handing over money for an app that demands arbitrary terminal access, he crafted an open-source alternative known as ituzov-gearbox.
The entire mechanism runs on top of tmux, targeting whichever pane hosts the running session of Claude Code. A local Node.js service intercepts the virtual lever movements and converts them into direct terminal keystrokes, firing off commands to switch active AI engines without requiring the developer to touch the keyboard.
To automatically discover available models without hardcoding names, the utility opens the assistant menu and scrapes the live screen text using native tmux capture commands. An internal regex guard checks whether the assistant is currently typing to ensure that closing the inspected menu with an automated Escape key does not abruptly terminate an active code generation cycle.
The transmission interface itself avoids bloated game engines, utilizing an H-pattern graph drawn in SVG and managed with vanilla JavaScript. The virtual knob is physically constrained to realistic gates and cannot clip diagonally across barriers, requiring shifts to pass cleanly through neutral while triggering mechanical audio clicks synthesized through the Web Audio API.
The shifter floats as a standalone borderless utility window powered by Chromium app mode with its own isolated profile, providing out-of-the-box support for Linux, macOS, and Windows environments running WSL.
Modern software engineering has finally arrived at the milestone where programmers spend entire weekends engineering manual transmissions just to avoid typing six characters into an AI prompt. The boundary between peak workflow customization and unhinged over-engineering has completely dissolved.
Source: GitHub
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.