I Tried Out OpenAI's New AI Keypad

OpenAI shipped its first consumer hardware product on July 15, 2026, and it is not the Jony Ive-designed ambient AI companion the industry has been speculating about for months. Instead, it is a $230 mechanical keypad called the Codex Micro, co-built with boutique peripheral maker Work Louder — and it sold out on day one.

The Claim

OpenAI is positioning the Codex Micro as a dedicated physical control surface for its Codex coding agent, sold through the company's new "Supply Co." storefront under a "Work Louder Co-Lab" label. The pitch is straightforward: instead of tabbing between windows and typing shortcuts, a developer can keep one hand on a purpose-built pad that manages their AI coding sessions directly.

The hardware itself is compact — 13 mechanical switches arranged around a rotary dial, a small analog joystick, and a touch sensor. The rotary dial adjusts Codex's reasoning level on the fly. The joystick triggers common coding workflows: reviewing pull requests, kicking off a debug run, or accepting and rejecting suggested changes. Six of the keys map directly to specific ChatGPT sessions or Codex agent contexts, meaning a single physical key equals a named agent. The pad ships with 32 interchangeable keycaps and connects via either Bluetooth or USB-C, with support for both Windows and Mac. Configuration lives inside a dedicated Micro tab within ChatGPT itself, where users assign brightness, key mappings, and project links.

The most visually distinctive feature is the RGB backlighting, and here OpenAI has done something worth noting. The lighting is not decoration — it is a status vocabulary. White means the agent is idle. Blue means it is actively reasoning. Green signals a completed task. Red flags an error. Multiple reports, including coverage from TechSpot and Slashdot aggregating an Axios story, confirm this color-coded scheme. OpenAI told TechCrunch to treat the Codex Micro as a limited-run project rather than a permanent product line, suggesting it will disappear once existing stock clears.

What We See

The honest read on the Codex Micro is that it solves a real problem for a very specific person — one who lives inside Codex all day, runs multiple agent sessions simultaneously, and finds the keyboard-and-mouse interface genuinely friction-heavy. For that narrow profile, the agent key concept has merit. Assigning a physical key to a named agent context is a concrete UX idea: you glance at the key color, you know the state of that session, and you push one button to interact. That is faster than reading a status sidebar.

The color-coded status vocabulary is the part I keep coming back to. The four-state model — idle, thinking, complete, error — maps almost exactly to what any multi-agent dashboard would need to surface. OpenAI has made a hardware argument for a software convention, which is an interesting move. Whether or not the Codex Micro itself succeeds as a product, that vocabulary now has a physical reference point, and developers building their own agent interfaces will recognize it.

What holds up less well is the broader positioning. The Codex Micro is described by TechCrunch as potentially "slightly mystifying" to anyone outside a specific Codex power-user cohort, and that framing is accurate. A voice dictation button and a send key — two of the pad's highlighted features — duplicate functions that already exist one or two keystrokes away on any standard keyboard. The rotary dial for adjusting reasoning level is a genuinely novel input for an AI tool, but it raises a question the available coverage does not answer: how often do working developers actually want to change reasoning depth mid-session, and does a physical dial improve that workflow over a dropdown?

Mashable notes the device includes features like the joystick for triggering pull request reviews and debugging workflows, but does not independently verify whether those integrations work smoothly in practice. The TechCrunch hands-on is the closest thing to a tested review across these sources, and even there, the conclusion reads as "interesting for coders, puzzling for most" rather than an unqualified endorsement.

Where It Falls Short

The price is the loudest objection, and it is a fair one. At $230, the Codex Micro costs more than many full mechanical keyboards. ZDNET's David Gewirtz tried to buy one when it launched on July 15 and found it sold out immediately — so he rebuilt its core functionality on a Stream Deck+, a general-purpose macro pad that many developers already own, and concluded it was a more customizable alternative at a lower price. TechSpot's headline called the Codex Micro "a Stream Deck for ChatGPT," which is both an accurate description and an implicit criticism: Stream Decks already exist, cost less, and work across every application rather than being locked to a single vendor's ecosystem.

That ecosystem lock is the more serious structural concern. The Codex Micro's configuration lives inside ChatGPT. Its status lighting is tied to OpenAI's agent runtime. Its value proposition depends entirely on continued investment in Codex and on the specific workflows OpenAI has chosen to surface through the pad. OpenAI itself acknowledged to TechCrunch that this is a limited-run item, meaning they are not committing to a hardware roadmap around it. For a $230 purchase, "buy it before it sells out and we may not make more" is not a reassuring durability argument.

There is also a hardware pipeline context worth naming. OpenAI's announced partnership with Jony Ive involves a screen-free AI device equipped with cameras and spatial awareness sensors — a fundamentally different product category, still unshipped. The Codex Micro appears to be an interim hardware move, not the company's primary bet on physical AI interfaces. Multiple sources place both efforts inside the same broader moment of OpenAI building out its consumer product surface, but the Codex Micro reads as a constrained experiment rather than a statement product. Whether it creates a repeatable hardware category, or disappears as a collector's item once the limited run ends, depends on whether OpenAI commits to the "Supply Co." concept beyond a single launch.

Sources