Codex Micro Brings Physical Controls to AI Workflows
OpenAI’s Dominik Kundel presents the Codex Micro, built with Work Louder, as a customizable hardware control surface for directing and monitoring Codex work. In a demo using GPT-5.6 Sol to build a browser word game, he uses dedicated controls for voice prompts, Plan mode, reasoning effort, task switching, permission approvals and queued follow-up requests. Kundel’s central claim is that users can map those actions to physical inputs around their own workflow rather than rely on a fixed keyboard layout.
Codex Micro maps active Codex work to hardware controls
Dominik Kundel presents Codex Micro as a customizable keyboard built with Work Louder for Codex. In his example, the device brings voice dictation, Plan mode, reasoning effort, task switching, permission approval, and configurable shortcuts into a set of physical controls while Codex is working.
Kundel uses GPT-5.6 Sol to build a browser game called “One Letter Off.” Players change one letter at a time to turn “cold” into “warm,” with a 30-second timer, streaks, and tile animations. Holding a key on the Micro starts voice dictation, transcribing the request directly into Codex’s “What should we build?” field.
I can hold this key and start talking to Codex.
The entire specification appears without Kundel touching the laptop keyboard. Later, he uses the same control to send a revision: “Make the win state punchier and add a streak counter.” In Kundel’s framing, the Micro is meant to support both the initial instruction and subsequent direction of work.
Input and steering are assigned to dedicated controls
The Micro’s analog stick can be mapped to the skills Kundel uses most often. Its default upward swipe activates Codex’s Plan mode; the interface confirms the command with “Toggle plan mode Triggered.”
A separate dial controls reasoning effort. Kundel says he uses it to move between a quick task and deeper reasoning without breaking his flow. The displayed Codex setting changes GPT-5.6 Sol’s advanced reasoning level from Light to Extra high.
The point is not that a particular mapping is mandatory. Kundel says the controls can be configured around the Codex actions a user uses most. But the defaults shown give distinct physical inputs to three forms of direction: speaking a request, entering Plan mode, and changing the requested reasoning effort.
Pinned tasks turn the Micro into a surface for supervising concurrent work
Kundel is not only building the word game. He also has other work running in pinned tasks, each assigned to a Micro key for quick switching. The visible Codex sidebar includes “Codex Chief of Staff,” “Docs revamp,” and “Build One Letter Off game.”
He says the device provides a quick view of what is running and what needs attention. When Codex is in the background, a double tap can surface an item requiring attention. Task switching, in this setup, is not limited to moving among open windows; it is linked to the pinned work Codex is carrying out.
The same control surface handles an interruption that requires approval. When Codex asks for permission to use Apple Notes, Kundel switches to the request and presses accept. The prompt shown asks, “Allow ChatGPT to use Notes?” and highlights “Allow this conversation.”
After approving the request, Kundel switches back and holds the voice key to queue the follow-up change to the game. He describes jumping among tasks as easier this way, adding that the analog stick also works as a fidget toy while Codex completes a task.
Customization is the product’s central premise
The completed game shows a timer at 23, COLD as the current word, WARM as the target word, and an instruction to type a four-letter word. Kundel calls it a live game built “all from the Micro.”
The particular controls used to make it are examples, not a fixed layout. Kundel says every Codex Micro control can be customized, including mappings for common skills, creating a pull request, or “even triggering my pet.”
All of the controls on the Codex Micro are customizable.
The settings screen shows battery and brightness controls, a graphical keyboard layout, agent keys and knobs, and an option to edit a keycap’s assigned shortcut. OpenAI positions the Micro as a configurable hardware layer for directing and monitoring Codex work: entering requests, selecting tasks, changing modes and reasoning effort, responding to permission prompts, and sending follow-up instructions.