I Drive Claude Code from a Kindle

I drive Claude Code from a Kindle. On the beach. In direct sunlight. No jailbreak.

Here's the thing about e-ink for AI: you don't need 120Hz to approve an agent's plan or read its output. You need a screen you can read in full sun, a battery that lasts weeks, and zero notifications to doomscroll into. It's the perfect human-in-the-loop screen.

So I put my Claude Code session on my Kindle, outside. The session itself runs in tmux on a small home server.

The catch: my Kindle 3's browser is from 2010. No WebGL, no working WebSocket. So every web terminal (ttyd, Blit, wetty) renders a blank page. They all stream over a WebSocket the Kindle simply can't do.

So I built a deliberately dumb one. A few hundred lines of Python, zero dependencies: render the tmux screen as plain HTML, send keystrokes with plain form posts. No WebSocket, no framework. Dumb on purpose, and that is exactly why it runs on that ancient browser. (I even bridged a Bluetooth keyboard through a phone, since the Kindle can't pair one itself.)

The "Claude Code on a Kindle" idea is having a moment right now. Most setups lean on jailbroken devices or local-model GPU fleets. This is the opposite: the smallest, most boring version. Runs on basically any e-reader, no jailbreak, no special hardware, behind your own auth. MIT, open source.

Repo + the whole story (yes, including the beach photo): github.com/smashbeat/claude-on-kindle

The best screen for this job turned out to be the slowest one I own, because it still works in the sun.