A sketch-and-voice canvas for AI
Draw it, explain it, send it to AI. Preppit keeps every note attached to the thing it's about — so the AI is told what you mean, instead of made to guess from a screenshot with three paragraphs typed underneath.
Press and hold anywhere to pin a typed note or a voice note to a spot, an object or a region. The note travels with what it's about, not just near it.
Hold K, say it all in one go, and touch the things you're talking about. Each thing you said becomes a pin on the thing you touched.
Before anything is sent you see the pass: what each note is about, what was read off the picture word for word, and what Preppit guessed rather than knew.
How it goes
Nothing is sent until you press Serve it and confirm what's on the pass.
A frame appears over whatever is under your cursor. Drag it around the part you mean — a slide, a chart, a paragraph, a bit of someone else's website. Or draw from nothing, inside the chat.
Circle the thing. Cross out the thing. Press and hold to pin a typed comment or record a voice note against that exact spot — transcribed, and you can fix the transcript.
Press the hotkey again and the next capture joins the first, in order. Lay the stack out as slides, a storyboard or plain pages, so a deck, a spreadsheet and a web page go across as one prompt.
Write one line about what you want. Preppit shows you what it's about to send — you confirm, and it lands in the conversation ready to work from.
Talk over it
Hold a key, talk the way you'd talk a colleague through it, and touch each thing as you mention it. When you let go, Preppit cuts the recording into notes at the pauses and pins each one to whatever you touched while saying it.
A note said while touching nothing is not guessed at. The pass asks you what it's about, with the nearest things as one-tap answers.
The x-ray
A screenshot is pixels until something reads it. Preppit reads the text off every picture before you send, and the x-ray shows you the result: green where it was sure, amber where it wasn't. Tap an amber line and type what it should say.
Green reaches the AI word for word. Amber may not — which is worth knowing before you send a plan that hangs on a number.
The pass
Serve it opens the pass: your line as the headline, every note with what it's attached to, what was read off the pictures, and anything Preppit needs to ask. Fix a note, answer a question, then serve.
The AI gets a clean picture, a numbered picture, and a brief that says in words what each note is about, in what order, and which parts are certain versus guessed.
A stack
A canvas becomes a stack the moment you add a second page. Lay it out as slides, a storyboard or plain pages; the text boxes read as headings, frames or labels to match. The AI gets a numbered contact sheet for the shape of it and can look at any page at full size.
Why bother
You can already paste a screenshot into a chat and type three paragraphs underneath it. Everybody does. The trouble is what happens to the relationships between what you drew and what you said.
Your arrows, circles and comments become pixels. The model has to guess which note goes with which thing, from how close they happen to sit. Small handwriting gets compressed. Order and emphasis are inferred, or lost.
The picture goes too — but so does a structured account of it: what you drew, what each note refers to, in what order, said in words rather than left in pixels. Where Preppit guessed rather than knew, it says so.
Honest about the claim. That this produces better answers than a screenshot and a good paragraph is the thing being tested, not a thing being asserted. In our own small tests so far, when the notes were pinned the AI worked out what every one of them was about; given the same notes as a list under the same picture, it worked out none. That's one result at a small scale, not a promise — it's why access is limited and why I want to hear what actually happens when you use it.
Setting it up
Email and a password, a link by email, or your Google account. Your work is yours: it isn't shared with anybody else's, and you can take it out or delete it whenever you like.
Your account page hands you one link per AI. In Claude: Settings → Connectors → Add custom connector, paste it, name it Preppit. Then say open Preppit in any chat and a canvas appears in the conversation.
It comes with your invitation. Sign in with the same account and it fetches everything else itself. This is the one that gives you the hotkey and the frame over your screen — the connector on its own gives you a canvas inside the chat, which is a good deal less fun.
The app isn't signed with an Apple certificate yet — a paperwork and money problem rather than a code one — so macOS blocks it the first time and words that block about as alarmingly as it could. You get past it once, and never see it again.
On macOS Sequoia and later it offers you only Done and Move to Trash. Click Done. Nothing is wrong.
Scroll to the bottom of that page. There's a line saying “Preppit” was blocked to protect your Mac, with an Open Anyway button beside it.
You'll be asked for your password or Touch ID, and then it opens. That's the last you'll hear of it.
On macOS Sonoma or earlier, it's simpler: right-click the app in Applications, choose Open, then Open again in the dialog. Apple removed that shortcut in Sequoia.
Or, from the Terminal, on any version — this strips the “downloaded from the internet” mark and the app then opens normally:
xattr -dr com.apple.quarantine /Applications/Preppit.app
What it needs, and what it sends
Where this is up to
Preppit works and is used daily by the person who built it. It is not a finished product, and you'd be an early user rather than a customer.
What I'd like back. Not a bug report. Whether you reached for it a second time, and whether what came back from the AI was closer to what you meant than typing it would have been. That's the whole question.
An account takes a minute. The connector link is on the page that follows, and the Mac app signs in with the same account — it makes its own link, so you never paste one.