USB Hardware for Coding Agents on Linux
Pedal Three ($69) is three ribbed foot pedals on a curved steel base with a captive USB cable. On Linux it is the easiest piece of hardware on this site to get working, for one reason: it is a keyboard as far as the kernel is concerned, and Linux has known what to do with keyboards for a very long time.
Do I need a driver?
No. A plain USB input device is claimed by the kernel's generic HID driver the moment you plug it in, and it appears as another input device alongside your keyboard. There is nothing to compile, no module to load, no package to install and no vendor daemon running in the background. Press a pedal and the character arrives in your terminal. That is the entire setup, and it is genuinely the reason a pedal or a keypad is a better bet on Linux than anything that needs an app.
Will the configuration software run on Linux?
This is the honest catch. Getting keys out of the device is free. Deciding which keys it sends is not always. Some devices are programmed from a web page, which works anywhere. Others are programmed by a small utility that the supplier only offers for Windows, and occasionally for macOS. Where that is the case the mapping is written into the device and stays there, so you configure it once on a borrowed machine and then use it on Linux forever, but you do need that one machine.
We would rather you knew that before ordering than after. Check the listing for the specific item, and if it is not stated, assume you may need to borrow a Windows machine once.
Does it matter whether I am on X11 or Wayland?
Not for the hardware. A pedal press is real input from a real device, so it reaches the focused window under either. It matters a great deal for anything that fakes input: a script that synthesises keystrokes, or a utility that grabs a global hotkey. Under X11 that has been solved for decades. Under Wayland, synthetic keystrokes go through the kernel's virtual input device, which your user needs permission to open, usually granted through a group membership or a rule file, and global shortcuts are the desktop environment's business rather than any application's.
The practical consequence is that a pedal sending a plain keystroke is the least fragile way to control an agent on Linux. Every layer you add above that is a layer that behaves differently on the two display servers.
Where does the hardware plug in if the agent runs on another machine?
Into the machine you are looking at, always. A keypad types into the focused window of the computer it is connected to, so if Claude Code or Gemini CLI is running on a small always-on box in a cupboard and you are driving it over SSH from a laptop, the pad belongs on the laptop. The same goes the other way for anything ambient: a light that tells you a run has finished is useless in the cupboard.
On the box itself, the things that matter are the dull ones. It stays on. It is quiet enough to live near you. It has enough memory to hold your toolchain and your editor at the same time. You can reach it over the network without ceremony. We do not sell computers and will not pretend to have an opinion on which one to buy.
What about dictation?
Linux is the weakest platform for this and we are not going to dress it up. There is no single system dictation service to bind a key to, so you are assembling one: a speech model or service of your choosing, plus something to type the result into the focused window, which runs straight into the Wayland limitation above. Our microphone hardware does not change that. The Whisper Key ($87.90) is a handheld microphone with a key on it, and it performs no recognition of its own.
If you want the simplest thing that works, start with Pedal Three.