How I built NotchWise

I've been using Anki for years and it was great for learning things, except for one problem. Every time I opened it to do a quick session, I'd lose twenty minutes to the app itself. The cards, yes, but also the settings, the deck browser, the urge to reorganise my tags. By the time I closed it, I'd forgotten what I was working on before.

That's the loop I wanted to break. Not to replace Anki, which is genuinely good, but to make studying something that happened around my work instead of pulling me out of it.

The notch turned out to be the right place

I tried a few things first. A menu bar app felt too hidden. A floating window felt too pushy. Keyboard shortcuts were fine but I kept forgetting them. Then one evening I was looking at the notch on my MacBook Pro, which had become a place where nothing useful lived, and it clicked. The notch is already in my peripheral vision. It's tolerated rather than attended to. If a card could quietly appear there, I wouldn't have to decide to go and look for it. It would just be there when my eyes drifted up.

For older Macs that don't have a notch, NotchWise draws a thin widget in the same spot. Same muscle memory, same rhythm, no one left out.

Eight seconds became the unit

Once the position was settled, the rest of the design came out of one constraint. The whole interaction needed to be fast enough that I could do it between paragraphs of real work. A card appears. I look up. I hover to flip it. I press 1, 2, 3, or 4 to rate it. It's gone. Total time, about eight seconds. If it ever took longer than that, I'd stop doing it.

That constraint shaped everything. The card layout is as spare as I could make it without losing information. The rating keys sit close to the home row so the motion stays small. The scheduler doesn't show me a card during a meeting or a focused coding block, because an interruption at the wrong moment is worse than no interruption at all.

The boring parts mattered most

A good chunk of the work went into things nobody sees. NotchWise runs entirely offline. No accounts, no cloud sync, no telemetry. Your data sits in a SQLite file on your Mac and that's it. It imports your existing Anki decks, both .apkg files and live AnkiConnect sessions, so you can switch without losing a single card. You can pick between SM-2, which Anki uses by default, and FSRS, the newer scheduler that's better at predicting what you'll actually remember.

The auto-hide logic was the thing I got most wrong the first few times. I shipped a build where cards kept appearing during screen shares. That afternoon I rewrote it to watch for ScreenCaptureKit sessions, Focus modes, and full-screen presentations. A flashcard about anatomy showing up in a Zoom call is the kind of moment you don't come back from.

What I didn't see coming

I'm working on a version that makes cards from whatever text you drag in, using Apple's NaturalLanguage framework so nothing leaves the machine. That part I expected to be hard. What I didn't expect was that I'd still be tweaking the scheduler weights every few weeks, months after launch.

Not because they're wrong. Because the line between a useful interruption and an annoying one is narrower than I thought, and it moves depending on what I'm working on. A card during a focused coding block is unwelcome. The same card during a half-finished email is fine. The system has to read context I haven't fully given it yet.

That's the part I'm still figuring out. NotchWise is on the Mac App Store if you want to try it, and there's a demo in the browser. Either way, I'd be curious to hear how the timing feels for you.