In my review of the reMarkable Paper Pure I said the journal would be useless if it stayed on the device. This is how it gets off. Each handwritten entry lands in Bear Notes as a dated, searchable note, transcribed by Claude, with the original page attached underneath.

It works now. Getting there taught me more about the device than three months of writing on it.

Two ways in, and neither is the advertised one

reMarkable wants you to use its cloud sync and its apps. For automation there are really two local paths, and neither needs a subscription:

• The desktop app’s local copy. The Mac app keeps your notebooks on disk. Reading them directly gives you the real structure: every page has its own ID, in order. A 37-page notebook reads in a fraction of a second.
• The USB web interface. Turn it on in the tablet’s settings, plug in, and the tablet serves a tiny website to your Mac. Ask it for a notebook and you get a PDF of the whole thing. That same 37-page notebook came down as 23.5 MB in 139 seconds.

The third option, logging into the tablet directly, is locked behind Developer Mode. So I use both local paths, and each one covers the other’s failures.

What broke

The USB interface turns itself off. I kept finding the setting switched off without having touched it. From the Mac, the tablet looks connected because it’s charging, but requests just hang until they time out. There’s no error to tell you anything is wrong. My first suspect was a macOS privacy permission. Granting it changed nothing. The fix was the toggle on the tablet.

The desktop copy sometimes won’t open. Reading it occasionally just stalls: no error, no activity, sometimes for many minutes. One morning it cleared on its own and the read finished almost instantly. Another time it stalled for about eighteen minutes and never came back. I still don’t know why, and I’m not going to pretend I do.

Long pages get silently cut off. This was the dangerous one. My journal uses the extended page, which keeps scrolling as you write. Export one of those the obvious way and you get only the top. In the worst case about seven-eighths of an entry was missing, and the export still reported success. The fragment looked like a complete short entry, and the transcription read like a finished one. Nothing flagged it. Now the pipeline checks how tall each page really is, renders the full height, and slices it into overlapping strips so no line is lost at a cut.

Pages move. A page I had already processed jumped from position 18 to 36 between two checks. Its ID stayed the same. So every page is tracked by ID, never by position. The USB path has no IDs at all, so when I fall back to it, the pipeline matches pages by their position and the date I write at the top of each entry, and that needs a reconciliation step afterward.

What lands in Bear

Each entry arrives as a note titled with the date, in both the Gregorian and Hijri calendars. The handwriting is transcribed, my headings and timestamps are kept, and sketches become short descriptions. Any word Claude isn’t sure of is marked with [?], so I can check it against the original page attached at the bottom. A footer records where the entry came from, for example: page 88 of my “Personal” notebook, extended page, ten slices.

That footer is the trust layer. When a transcription reads oddly, I can see exactly which page it came from and what was done to it.

The lesson

The tablet is built for one person on one device, and it’s excellent at that. Everything I needed to get my writing back out was possible without the cloud. The hard part wasn’t the AI. It was the boring stuff: a setting that switches itself off, a read that stalls, a page that’s taller than it looks.

The transcription is the easy part. Making sure there’s a whole page to transcribe is most of the work.