Shot when you speak
The spot under your pointer is captured the moment you start talking, and again when you pick up after a pause. Up to 3 screenshots per note, each with a quote of what you said. Outlined and dimmed, never drawn over.
Mac · Windows · Linux
Works with ChatGPT, Claude, Gemini
Point at it. Mumble about it. Go back to bed.
Lazy as turns your cursor and your voice into screenshot-annotated notes, then squeezes them into a tidy prompt for ChatGPT, Claude, Gemini or Copilot. For people who'd rather point than type.
Illustration: a cursor rests on a pricing page button and a red outline snaps around it as it's photographed. The reviewer talks, says "next note", and note 1 is saved with its screenshot and a quote of what they said. Then a rambling 1,650-token set of notes becomes a 480-token prompt.
Made for reviewing
- Designs
- Websites
- Videos
- PDFs
- Claude artifacts
- Staging builds
- Slide decks
The film · 1:01
Watch a cursor quit its job.
Sixty-one seconds on what Lazy as does, set to a beat. Turn the sound on. It took more effort than the app wants you to make.
How it works
Three steps. One of them is talking, which you already do too much.
-
1
⌃Z
Press the shortcut
Ctrl+Alt+Z on Windows and Linux. Or hold the middle mouse button for half a second on a Mac, or flick a Logitech MX gesture. A sleepy little HUD says it's listening.
-
2
pointresttalk"next note"
Point, rest, talk
Put the pointer on what's wrong and say why. It's photographed as you start talking, and again each time you start after a pause, with a quote of what you said. Pause as long as you like. Say "next note" (or press ⌃X) when you're done with that one.
-
3
⌃Z⌃⌥Z
Stop, polish, paste
Press ⌃Z again and your notes are on the clipboard. ⌃⌥Z turns them into a compact prompt. Paste it into ChatGPT, Claude, Gemini or your coding agent. Nap.
Features
Your cursor already knows what's wrong. Now it can tell your AI.
Everything a reviewer needs, nothing that requires sitting up straight.
Snip mode ⌃⇧Z
Drag over the exact area like ⌘⇧4 (Ctrl+Alt+S on Windows and Linux), then talk about it. For when "around here" isn't precise enough.
Voice commands
Notes end only when you say "next note" (or press ⌃X), so you can think mid-sentence. Say "general comment" for feedback about the whole thing, with no screenshot. "Scratch that" drops the note you're on.
Speaks your language
Any language Apple's speech recognition supports on a Mac, and Whisper on Windows and Linux. Got two? Press ⌃L to flip mid-session. Long monologues are kept in full, every word.
Remembers the page
Records the page with each note (the browser address on a Mac, the window title on Windows and Linux), so your AI knows which page you meant.
Reuses screenshots
Three notes about one button? One image. Fewer files, fewer tokens, same amount of complaining.
Mouse and Logitech friendly
Hold the wheel button for half a second on a Mac, or map an MX Master gesture to the shortcut. Hands never leave the mouse. Barely leave the couch.
Works over full-screen apps
Browsers, Figma, video players, PDF viewers, full-screen or not. If you can point at it, you can complain about it.
Naps on its own
Forgot to stop? It stops itself after 5 quiet minutes (or whatever you pick). The mic deserves a break too.
Token diet
AI reads by the token. Your rambling doesn't have to.
"Make it a prompt" rewrites your umms and self-corrections into a short, lossless instruction. Every request, value and quote stays. The filler goes. In our example, 71% fewer tokens.
- Keeps every screenshot path, so coding agents like Claude Code and Codex read the images locally.
- Images are never sent while polishing. Only the text of your notes.
- Optionally writes the prompt in English, whatever you spoke. Fewer tokens again.
Example 12-note session. Your mileage depends on how much you ramble.
Screenshot sizes
| Setting | Per screenshot | Good for |
|---|---|---|
| Light default | ~500 | Most reviews |
| Balanced | ~800 | Busier pages |
| Sharp | ~2,000 | Tiny text, dense UIs |
Privacy
Too lazy to spy on you. Also, we just wouldn't.
No passwords, no analytics, nothing of yours in our cloud. Your screenshots and notes live in a folder on your computer.
Stays on your computer
Screenshots, transcripts and notes are saved in Documents/LazyAs. Audio is never recorded to disk.
Speech, your call
On a Mac, Apple's speech service by default, like Dictation, or Private to keep it on your Mac. On Windows and Linux it never leaves your PC.
Text only, on request
Only the polish step sends note text to the AI you picked, and only when you click it (or turn on auto-polish).
No passwords here
You sign in to your AI in its own official app, not in Lazy as. An optional API key sits in the macOS Keychain. Not in a text file named secrets_final_v2.
It's less effort than reading this far.
Download it, press ⌃Z (Ctrl+Alt+Z on Windows and Linux), complain at your screen. Your AI takes it from there.