ACID CAT

Every month brings another agent framework promising autonomous workflows, and they all break in the same place. Not on reasoning, and not on the model. They break at the interface, the moment the software they are driving decides not to cooperate.

So the question is less about intelligence than about access. What would it take for an agent to use a computer the way a person does, without asking any application for permission first?

Every existing answer is software asking software for cooperation. An API can be rate limited or simply may not expose the thing you need. A browser extension depends on the browser. A driver needs install rights on a machine you may not control. Each layer is one more party that can say no, change shape, or disappear in the next release.

Going underneath all of that means being real input hardware, and that used to be a serious undertaking: firmware, USB descriptors, a board, an enclosure, and a manufacturing run. Historically that meant a hardware team, which is why the agent world stayed in software.

Stop integrating and start typing. ACID CAT is a small device that plugs into a USB port and presents itself as an ordinary keyboard and mouse. There is no driver, no extension, and nothing to integrate. To the operating system it is simply input, which means it works in any application, on any operating system, including the ones that were never going to publish an API.

The insight is that the human interface is the one integration surface that every piece of software already supports and none of them can withdraw. It has been stable for decades precisely because everything depends on it.

• Microcontrollers with native USB. A few dollars now buys a chip that can enumerate as a compliant HID device. The hard part of the old version of this project is now a library call.
• Cheap fabrication. Small boards in days for pocket money, enclosures printed overnight. A physical product can be iterated on a coffee budget rather than a tooling budget.
• A thirty-year-old standard. USB HID is boring, universal, and finished. Building on something nobody is going to redesign is its own kind of leverage.
• Direct sales. A landing page and a payment link replace a distributor, so a small batch can be sold to the people who want it without a retail channel.

ACID CAT v1.0: a small USB device that acts as keyboard and mouse for whatever is driving it, plus a tiny LCD on the front where a bongo cat drums along with every keystroke. That last part serves no technical purpose at all, which is precisely the point of building your own hardware.

It is on sale as a limited first batch at $75, with checkout handled through Stripe and units shipping in the order they were reserved.

• Hardware keeps its own schedule. The landing page went up before the product photography did, which is an honest signal about where the project actually is: the design is settled, the first batch is not yet in people’s hands.
• “It is just a keyboard” is true right up until you want structured feedback. A device that only sends input is blind by design, so anything that needs to read the screen still lives on the other side of the connection.
• Selling a physical object turns out to be a different discipline from building one. Shipping, batching, and telling people honestly when their unit will arrive are their own work, and none of it is firmware.

A designed product with a price, a checkout, and a first batch open for reservation, taken there by one person in the time a hardware company would still be scheduling reviews. It has not shipped yet, and saying otherwise would be the easiest lie on this website. Pre-orders are open at acidcat.dev; the honest status is that the interesting engineering is done and the logistics are next.

If a single person can take a physical product from a question to a purchasable object this quickly, then hardware stops being the part of an idea you abandon. That matters well beyond this device: instruments, field tools, one-customer machines, and all the small objects that were never worth a manufacturer’s attention become reasonable things for a tiny team to make.

Notebook The most durable interface in computing is the one every piece of software already has to accept.