In just a few short years, we've gone from marveling at chatbots that could hold a convincing conversation to agents and coding tools used to accelerate software development throughout the tech world. AI power users have embraced vibe coding and multi-agent workflows, and those evolving tools call for evolving interfaces. Here comes the Codex Micro ($230), a dedicated keypad made for the vibe coder era, with programmable macro keys and built-in functions to seamlessly coordinate multiple AI workflows and agentic tools at once.
Born from a collaboration between OpenAI and Work Louder, a peripheral maker best known for making retrofuturistic mechanical keyboards, the Codex Micro is a curious device. It bears OpenAI's logo and works with the ChatGPT desktop app, but it's essentially an off-the-shelf product with some special flourishes in both the hardware and software.
Speaking of rapid evolution, this hardware comes as OpenAI's Codex agent ecosystem itself is changing. OpenAI recently launched ChatGPT Work, a task-completion agent, and merged the standalone Codex app into the ChatGPT desktop app. Now, all three—Chat, Work, and Codex—are part of the same shared ecosystem.
Plenty of people scratched their heads when the Codex Micro was announced. (What's it for? Who asked for this?) But the limited-run macro pad sold out in less than a day. I got my hands on one, and after a couple of days of use, I have some distinct first impressions.
Codex Micro for Macros: What Exactly Does an AI Keypad Do? #
First, I should clarify that this may be OpenAI's first branded piece of hardware, but it's not the Jony Ive-designed smart speaker that has been rumored for months. Instead, OpenAI made this product as part of a licensing partnership with boutique manufacturer Work Louder. It's a reskinned version of Work Louder's Framer Micro and Creator Micro pads, though based on the manufacturer's Pro versions that include both Bluetooth and wired USB-C connectivity out of the box. It works with Apple macOS, Microsoft Windows, and Linux, so developers of all flavors can use it.
But why would people *want *this device?
The premise behind a macro pad is simple: It's a collection of dedicated keys, separate from your regular keyboard, exclusively for shortcuts—executable key combinations, typed text strings, or even multi-step automations, each tied to a single button.
I'm already a macro pad user. I have a Razer Nostromo keypad on my desk, and it's been a part of my daily workflow for years. I use macro commands for things like adjusting text formatting as I write, triggering a one-button option to type out my shipping address and info, and speeding up the little one-off actions I do most often. It makes sense to me that a developer who spends long stretches of time running the same commands in ChatGPT and Codex would want a macro pad, especially with added agent status lights and a dedicated voice button.
Perhaps the best way to think about the Codex Micro is in relation to the Copilot key on Windows laptops. What used to be just a one-button trigger to call up an app has evolved into a multi-button, multi-input device for supervising and orchestrating multiple AI tools and agents at once. So, this isn't about replacing your keyboard: It's an additional, faster input that lets you home in on the commands you use most frequently and gives you a way to monitor multiple streams of work at once. That's where the real utility comes in.
Status Lights for Keeping Tabs on Your Agents #
AI agents, especially coding agents like OpenAI's Codex, have a whole language of functionality that coders are familiar with, but the rest of the public may not know. Beyond simple prompts and responses, you'll find options to fork and merge different projects, accept or reject code suggestions, upload files, push files to the cloud, switch modes or models, and more. A programmable macro pad already has some appeal for a professional using Codex every day.
However, when you're using Codex or ChatGPT as an agentic tool, with tasks running in the background, switching between multiple projects makes it a headache all its own to keep all those ongoing threads straight.
This is why the Codex Micro has six translucent buttons dedicated to pulling up your most recent threads, each with a multicolor LED that indicates the status of the coding agent as it works on your tasks. Blue is for thinking, green is for a completed task, amber is for items needing additional input or feedback, red is for errors, and solid white is for an idle agent.
Even the clear polycarbonate border of the macro pad lights up to signal voice/mic status, with a sea-green color to show that the mic is live and recording, a white snaking effect to show that Codex is processing your voice prompt, or a simple solid-white glow to show that the prompt has processed fully.
The LED patterns and colors used for signaling are easy to distinguish, so you can keep track of what it's telling you instead of constantly checking your monitor or switching windows.
That's a trick I haven't seen any other macro pad do, from my trusty Nostromo to the latest Elgato Stream Deck. Sure, you could probably find a way to implement similar agent-status signaling using the Stream Deck's OLED buttons. Also, watching the status lights change may cause a little bit of secondary anxiety, becoming just something else to pay attention to, rather than removing one. At any rate, the feature seems genuinely useful.
The Codex Micro's other unique feature is a dedicated microphone button for voice prompts. While you could probably configure something similar (or task your Codex agent to code it for you), this ready-made device is a simple plug-and-play solution. The only quirk: The macro pad doesn't actually have a microphone, so it triggers your mic—provided you have one on your laptop or headset.
Heavy Metal: Clicky Switches and Milled Aluminum #
The Codex Micro's packaging and presentation are excellent—a premium unboxing experience on par with Apple and other high-end products. The product itself is quite heavy and substantial. In fact, at 2.6 pounds, it's way heavier than my trusty, 0.65-pound Razer Nostromo keypad, despite being half the size and with fewer keys.
The milled aluminum base's density and weight mean it will stay in place on your desk. It has a grippy rubber foot as well for traction, so it won't shift around unwanted. Material-wise, the frame is transparent CNC polycarbonate plastic with a sandblasted, anodized aluminum base. Sandwiched inside that stack is the actual circuit board and power supply.
The mechanical key switches it uses are pleasing in a clicky, tactile sense. They feel crisp and satisfying when you click, with generous key travel, and you'll feel a distinctive feedback bump when you press each key. It is a little loud, so I'm glad that Work Louder sells a non-clicky version, which would be better suited to shared offices.
This keypad sits at a slight slant rather than flat on the desk, which is comfortable to use and more ergonomically helpful for your wrists. It sits no higher than the SteelSeries mechanical keyboard I use, except for the knob dial in the upper-left corner—that one sits a little high.
The keypad itself has 12 mechanical buttons, one clickable knob in the upper-left corner, and a planar joystick in the upper-right corner. The knob twists and depresses like a button (with programmable options for both a short and long press). At the same time, the planar joystick—a grippy disk that slides around for directional control—provides a fidget-friendly tool for navigation, or you can program the individual up/down/right/left directions with specific macros, AI skills, or settings controls.
In fact, every button and input is customizable to a different function. With more than a dozen buttons, plus the knob and joystick, and multi-layer controls (that one keypad supports six layers), this one keypad can handle 120 different programmable actions, and only the first layer of controls is preconfigured out of the box.
Cryptic Icons and YOLO Mode #
The Codex Micro arrived in two boxes: the primary device and a box of 31 additional keycaps, for a total of 32 icons and symbols. Popping one off and putting a new one on is straightforward, though you might panic a little the first time you do it because it feels like you might break something.
You can find a full list of shortcuts assigned to these icons in the Codex Micro app interface. Most of them are pretty straightforward, though some are tied to functions that can be a little inscrutable to non-developers. If you're familiar with GitHub, then you might be more conversant in details like code reviews, pushing commits, and sending something for feedback. You'll see a checkmark to approve, an X to reject, a lightning bolt to enable fast mode, a terminal prompt to open the terminal panel, and a little bug to enable automated debugging functions. You'll also find a number of blank caps that you can assign to any shortcut you want.
Then again, some of these are even more inscrutable or just redundant. For example, you'll notice three different icons for toggle review (showing or hiding the review for a current Git-backed chat). Some don't seem tied to the obvious icon meanings to me, such as a paintbrush for adding photos to the active composer, or a party popper shooting confetti as the icon for opening a side chat.
Speaking of redundancy, the keycaps include two different little brain icons—one with more wrinkles and one with fewer—used to increase or decrease the reasoning effort in your current composer panel. However, you can also program the knob to serve as a reasoning-level selector, letting you control compute and token usage by turning the reasoning up or down. This redundancy speaks to the customizability of the controls, letting you choose between a fidgety knob or discrete buttons to fine-tune the model and reasoning selections as needed. (This particular feature is also a huge boon to AI developers, since it lets them more easily rein in token use and higher bills.)
The most interesting options are the text caps with "YOLO" and "YEET" icons, which trigger specific actions within the Codex composer: YOLO triggers a full-automation mode that can bypass manual user approvals entirely. (The actual CLI command is "--dangerously-bypass-approvals-and-sandbox.") Meanwhile, YEET stages all your local changes, then commits and pushes them to your Git repository. It then automatically opens and populates a pull request on GitHub explaining the changes—literally "throwing the code" into GitHub and letting Codex handle everything.
Setup Friction: Not Quite Plug-and-Play #
My biggest complaint about the Codex Micro is actually that it's not quite as plug-and-play as I would like. Mac and Windows are both supported out of the box, and while Linux is supported, it is maintained by the community rather than by Work Louder or OpenAI. Furthermore, while you can connect it to Android and iOS devices, configuring it and managing setup requires a desktop operating system.
The device plugs in just fine, and you can access controls and programs for it through the ChatGPT app, but it's not intuitive to do so. Switching between apps is a pain, but the real difficulty is in managing the wealth of control options that are available. Among preprogrammed Codex skills, chat controls, navigation options, commands for settings, the side panel, and the ChatGPT app itself, the sheer number of ready controls is a bit overwhelming.
The user interface in the Codex and ChatGPT apps is simple, but when you need to do more complicated, non-Codex key bindings, you have to use Work Louder's standalone Input app. Even with my prior experience programming macros, the overall process just isn't as straightforward as I would have expected, and most of that friction came from the bifurcated software app—half in ChatGPT, and half in the Input app.
That's the real disappointment here. So much of what the product promises depends on OpenAI's software side, but the execution stumbles by scattering controls across multiple places.
Tangible Tools for Invisible Tech #
In practice, using the Codex Micro feels like a cross between a macro pad and a mixing board. The programmable keys are easy to map and use, the LED status lights are a clever way of managing several projects at once, and the additional knob and joystick add tactile controls that (when set up for your needs) are intuitive. I'm not an AI developer by any means, but I could see these dedicated controls being a massive time saver for power users.
No, it's not the long-awaited Jony Ive device, but Codex Micro is a limited-run product with OpenAI branding and unique thread-tracking functionality. In the tech-obsessed corners of the market, that extra Silicon Valley cachet is enough to make this something of a collector's item. (Resellers are already hawking the thing on eBay for up to $1,000.) For reasons both practical and impractical, OpenAI's macro pad experiment is already a success, having sold out immediately.
Whether it signals a permanent new wave of AI-tailored desktop hardware remains to be seen. As a macro user, I see the utility, especially when it includes a visual dashboard for managing multiple agentic workloads. This new niche suggests making our invisible AI tools tangible is a workflow step worth taking. I just think it needed a little more polish before its release into the wild.
Disclosure: Ziff Davis, PCMag's parent company, filed a lawsuit against OpenAI in April 2025, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.