Questions, asked lazily

A review app for Mac, Windows and Linux that lives in your menu bar or tray. Press a shortcut, point at things on screen and talk. Each time you start talking (and again after a pause), the spot under your pointer gets a cropped screenshot with a quote of what you said, plus your full transcript and the page, saved as Markdown. Then it can turn the whole thing into a compact prompt for ChatGPT, Claude, Gemini or Copilot.

Anyone who reviews things and hands the feedback to an AI chat or coding agent: designers, developers, founders, PMs, video editors. If you've ever typed "the button on the left, no, the other left" into a chat, it's for you.

Everything you said until you said "next note" (pauses included), the app or browser page you were on, and up to 3 cropped screenshots of the spots you rested the pointer on longest, each with a short quote of what you were saying at that moment. See a real notes.md example.

Because people think mid-sentence. A note now ends only when you say "next note", press ⌃X, or snip. The HUD pill flashes "#3 saved. Now on #4." so you know it worked. Want silence to split notes anyway? Settings › Capture › Also start a new note after a silence of 4, 6 or 10 seconds.

Each session is named after the page you reviewed and the things you talked about most, and your AI gives it a sharper title when you make a prompt. In Review sessions…, click the name at the top to rename it. To delete a whole session (notes, screenshots and prompt), click the trash button next to the gear, right-click it in the list, or press Delete. It goes to the Trash, so you can still put it back.

Press ⌃Z (Ctrl+Alt+Z on Windows and Linux), or on a Mac hold the middle mouse button for half a second, or use a Logitech MX gesture. Same thing again to stop. Details in Shortcuts and Mouse and Logitech.

Almost certainly: Arabic, French, Spanish, German, Hindi, Chinese and many more. On a Mac it uses Apple's speech recognition, on Windows and Linux it uses Whisper (about 100 languages). It listens in one language at a time (more accurate). Set a quick-switch language and press ⌃L while recording to flip between the two. See Languages.

Macs with macOS 14 Sonoma or later (Apple Silicon and Intel), 64-bit Windows 10 and 11, and 64-bit Linux with an X11 desktop (Ubuntu 22.04 or newer, or similar). See the download page.

Yes. Say "general comment" (or press ⌃G, Ctrl+Alt+G on Windows and Linux) and talk. That note gets no screenshot, and the prompt lists it first, under General. Starting a note with "General," and a short pause works too. See General comments.

Same notes, same voice commands, same AI choices and the same session folders, so sessions move freely between computers. The differences: shortcuts add Alt, speech is turned into text on your PC with Whisper, and the page is recorded as the window title. The details.

You're probably on Wayland, which doesn't let apps capture other windows or listen for global shortcuts. Log in with an X11 session instead (on Ubuntu: "Ubuntu on Xorg" from the gear on the login screen). Install on Linux.

It listens while a session is running so it can transcribe you, but audio is never saved to disk. Only the transcripts and notes are saved.

On a Mac, by default (Transcription: Best accuracy) it's processed by Apple's speech service, just like macOS Dictation. Choose Private in Settings › Capture to keep it on your Mac, for languages marked ⚡︎. On Windows and Linux it never leaves your PC: Whisper runs right there. Your speech never reaches our servers.

Only when you make a prompt (or turn on auto-polish): the text of your notes, meaning screenshot file names, page names and what you said, sent to the AI you picked through its own app. Never the images. Nothing is sent while you're recording. Open in AI chat only opens the chat and fills your clipboard: you decide what to send.

The app has neither. There's nothing to sign up for: if you buy, your email at checkout is your account, and your licence signs you in with an emailed link. The app only calls home to check for updates once a day and, with a licence or trial, to renew its activation about once a week (exactly what it sends). This website does count visits and downloads, so we know how people find it. You sign in to your AI in its own official app, and an optional API key or Claude token lives in your macOS Keychain. Full details in the privacy policy.

No. Taking notes needs no AI at all, and Open in AI chat works with any chat on any plan, free ones included. To have Lazy as write the prompt itself, use a plan you may already have: a free Google account (Gemini, through Antigravity CLI), GitHub Copilot Free or any Copilot plan, ChatGPT Plus or higher (through Codex), or Claude Pro or Max (through Claude Code). Or use the free offline cleaner. Choose your AI.

Most people don't have one, and many basic plans don't include API access. Lazy as runs through each AI's official app on your own account instead, so it uses the plan you already pay for, or the free one. On a Mac you can still paste an Anthropic API key if you prefer paying per use.

Making a prompt cut the text by 71% in our example (a rambling 1,650-token session became 480), without dropping any request, value or quote. Screenshots are cropped so each costs about 500 tokens on the default Light size (around 800 at most).

Barely. One polish is usually 1 to 2 thousand tokens, run as a tiny request with no tools and no history. It counts toward your plan like any other use.

Yes. With Claude, Haiku 5.5 is the default, Sonnet 5.5 and Opus 5.5 are one click away, and you can type any model ID. ChatGPT, Gemini and Copilot use their app's default unless you type a model in. See Choose your AI.

No. Lazy as is an independent app that works with ChatGPT, Claude, Gemini and Copilot. It isn't made by, endorsed by or affiliated with OpenAI, Anthropic, Google, GitHub or Microsoft. Their product names are their trademarks.

In Settings › Setup click Fix next to Screen Recording, turn Lazy as on again, then click Relaunch. macOS remembers the exact build it approved, so an update can leave a stale switch. Full steps.

Yes, and that's the best setup. Lazy as listens through your Mac's own microphone whenever AirPods or other Bluetooth headphones (or your iPhone) are the Mac's mic, so your headphones keep playing in full quality and nothing cuts out mid-note. You can pick a different mic in Settings › Capture › Microphone. Older versions used the AirPods' mic, which could leave notes empty: update to 1.5.

When you start talking in a note, and again each time you start talking after a pause of 3 seconds or more (2 or 5 in Settings › Capture). It's taken where the pointer is. If you start talking and then move to the thing, or scroll, and settle within a moment, it photographs where you settled. Talking straight through? One screenshot. Need an exact area? Snip with ⌃⇧Z.

You're on speakers, so it heard the video too. Wear headphones when reviewing anything with sound.

On notched MacBooks a busy menu bar can hide it. Hold ⌘ and drag icons to make room, or open Lazy as from Spotlight to get to Settings. More.

Settings › AI › Sign in again (for Claude on a Mac: Connect my Claude). Until then, prompts fall back to the offline cleaner. Details.

That's the Keychain checking on your saved API key or Claude token. Click Always Allow and it won't ask again until the next update.

That was a bug in versions before 1.3 with long notes. Update to the latest version: the whole session is now transcribed continuously and long notes are kept in full. Details.

Lazy as is free during early access. If an AI writes your prompts, that usage counts toward your own plan with that AI (or your API account), not anything from us.

This early build isn't notarized by Apple yet. Click Done, then System Settings › Privacy & Security › Open Anyway. The install guide walks through it.

The early Windows build isn't code-signed yet, so SmartScreen doesn't recognize it. Click More info, then Run anyway. Install on Windows.

Download the latest version from the download page: on a Mac, replace the app in Applications; on Windows, run the new installer. Your settings and sessions stay. Check the changelog for what's new, and Support for bugs and ideas.