FIELD NOTES / EXPERIMENT 001
A notebook
that writes
back.
A pen. A paper tablet. A thought.
And then, an answer appears in the ink.
We gave a reMarkable a direct line to Gemini.
It feels a little impossible. Here’s how it works.
a very new feeling.
Why is the sky blue?
remarkable.
reMarkable 2
+Gemini 3.8 Flash
+Wi-Fi
=a conversation on paper
01 / THE EXPERIENCE
Think with your pen.
The page will keep up.
Write naturally. Touch with four fingers.
Lift your hand, and let the page answer.
Why is the sky blue?
A little room for an answer…
THE MOMENT IT CLICKED
“unplugged and
working great!”
That’s the whole point. The assistant lives on the tablet.
The USB cable helped us build it. The laptop gets to sleep.
02 / UNDER THE COVER
Follow the ink.
A small chain of ordinary technology
makes an extraordinary feeling.
The gesture tells the assistant to capture the visible page. Your handwriting, drawings, and earlier replies on that page become the context for this question.
01 / 04The tablet does the capturing and drawing. Google runs the model. The laptop isn’t in this loop.
03 / THE BUILD NOTES
A little less wizardry.
A little more tinkering.
The idea was simple. Making it feel simple took some work.
- 01
Start with a good idea.
Smart Remarkable and Ghostwriter showed the way: see what’s on the page, ask a vision model, then write into the notebook.
- 02
Teach it this particular tablet.
We built for the reMarkable 2’s ARM processor and adapted screen capture to its firmware. Now Gemini sees the right page.
- 03
Make the answer feel like it belongs.
Larger letters. More breathing room. A consistent fineliner. Replies arrive line by line, and the original writing stays.
- 04
Pull the plug.
The assistant starts on the tablet and calls Gemini over Wi-Fi. The last test was the best one: unplug USB and keep writing.
Still drawn ink. Just a lot easier to read.
04 / A FEW GOOD QUESTIONS
So… is it magic?
Mostly software.
A little suspension of disbelief.
Does the laptop really get to be off?
Yes. The assistant runs on the reMarkable and talks directly to Gemini using the tablet’s Wi-Fi. USB was used for setup and testing. The tablet still needs an internet connection to get a new answer.
What does Google actually receive?
An image of the visible page, along with the assistant’s instructions. That includes whatever notes and earlier replies are on screen. Requests use a Google API key and its project’s quota and billing, separately from a Gemini chat subscription.
Can I use any page?
The drawn-reply workflow works in notebooks and on PDF pages. Leave blank space for the reply and keep the page still while it answers. It sees the current view, so keeping the earlier conversation visible helps with follow-ups.
Does it know my entire notebook?
No. The current implementation sends the visible page. It doesn’t automatically search the rest of your library or remember conversations on other pages. The page itself is the conversation history.
Is the answer typed, drawn, or streamed?
Drawn. We tried native typed text, but it wasn’t reliable in everyday use. The working version traces readable letter shapes as editable ink. Lines appear progressively after Gemini finishes generating its response; this isn’t live token streaming.
What is running under the hood?
This build uses a reMarkable 2 on firmware 3.27.3.0, a customized Rust assistant, and Gemini 3.8 Flash with medium thinking. A four-finger gesture triggers it. A background service starts the assistant on boot. Firmware compatibility is specific to the device we tested.
What about websites and a daily newspaper?
We explored article delivery and a daily paper, then paused them to focus on the notebook experience. Those companion tools need a reachable computer. They’re separate from the untethered Gemini conversation shown here.
A SMALL EXPERIMENT IN QUIETER COMPUTING