Converting a typewriter into a Claude terminal

An analog AI interface

I converted a typewriter into a physical AI terminal.

When you type in a question, Claude will type back a response.

This was primarily a hardware hack. To allow Claude to type, I had to electronically simulate phantom keypresses. This meant fully intercepting the keyboard signals and re-routing them to be processed by an Arduino.

Like all keyboards, the typewriter keys were wired in a matrix. When a key is pressed, it connects a row to a column. By intercepting all of the 8x rows and 8x columns, I could scan the keyboard for keypresses and send back “phantom typing” at will.

This was slightly fiddly with 32 pins to tap and no space inside the enclosure. I added the electronics onto the back, including a Raspberry Pi Zero as the brains.

Claude’s physical form

It’s also a full linux terminal, since the intercepted keypresses feed directly into the Raspberry Pi over serial.¹ This is pretty fun to play with.

For some reason, seeing an LLM physically manifested is quite compelling. Watching something intelligent whir and clunk into life - answering anything you throw at it - felt like rediscovering LLMs for the first time.

If you’re in London and would like to have a go, drop me a note :)

1. I was inspired by Artillect’s Linux terminal typewriter and used a similar hardware approach. I’m very thankful for his documentation which saved me loads of time!