> ## Documentation Index > Fetch the complete documentation index at: https://keebye.com/llms.txt > Use this file to discover all available pages before exploring further. # Install and permissions > System requirements for Mac and Windows, the onboarding window, each permission Keebye asks for, updates, and where models are stored. Keebye ships as a signed and notarized DMG for Apple Silicon Macs and as an installer (.exe or .msi) for 64-bit Windows 10 and 11. This page lists what it needs from your computer and what each permission does. ## Requirements - Mac: macOS 11.0 or later. - Mac: an Apple Silicon Mac. The Mac release is built for arm64 only. - Windows: Windows 10 or 11 on a 64-bit (x64) PC. - Room for the default speech model: about 670 MB on disk, downloaded once. ## Installing on Windows Run Keebye__x64-setup.exe, or the .msi if you install software from MSI packages. Keebye then starts in the system tray and opens the same onboarding window as on a Mac. > **Warning:** The Windows installer is not code-signed yet, so Windows SmartScreen may say "Windows protected your PC". Choose More info, then Run anyway — only for an installer you downloaded from keebye.com. ## The onboarding window The "Welcome to Keebye" window shows five steps. "Finish unlocks once the model is downloaded and both permissions are granted." On Windows both permission steps show as granted, so Finish waits only for the model. The same window reopens later as a repair surface if a permission goes missing, for example after an update. 1. Downloading the speech model — about 670 MB, one time. This runs while you finish the steps below. 2. Microphone access — press "Grant microphone access". 3. Accessibility access — on a Mac, press "Open System Settings" and switch Keebye on. 4. Pick your push-to-talk key — Fn / Globe, Left ⌘, Right ⌘, Left ⌥, or Right ⌥ (Windows equivalents are listed in Getting started). 5. Try it — hold your hotkey and speak into the test field. ## Microphone On a Mac, macOS asks: "Keebye listens to your microphone only while you hold your dictation hotkey, to turn your speech into text." Keebye requests this in onboarding step 2. On Windows there is no prompt from Keebye; Windows grants microphone access when recording starts. > **Note:** No prompt and not turning green on a Mac? Open System Settings ▸ Privacy & Security ▸ Microphone and switch Keebye on there. On Windows, if dictation records nothing, open Settings ▸ Privacy & security ▸ Microphone and allow desktop apps to use the microphone. ## Accessibility (Mac) On a Mac, Accessibility access lets Keebye watch your hotkey and paste text into any app. Keebye polls the grant every 1.5 seconds and re-arms the hotkey every 2 seconds once granted, so you do not need to restart. Windows needs no equivalent permission. > **Warning:** Already listed in System Settings but not turning green here? Remove Keebye from the Accessibility list with the − button, then add it back and switch it on. ## Speech Recognition (Mac) On a Mac, requested only when you select the "Native (Apple Speech)" engine. macOS shows: "Keebye uses on-device speech recognition to turn your dictation into text. Audio stays on your Mac." Until granted, the pill shows "Speech Recognition permission needed — enable Keebye under System Settings › Privacy & Security › Speech Recognition". > **Note:** Input Monitoring is not requested. On a Mac, Keebye needs only Microphone and Accessibility, plus Speech Recognition for the Native engine. ## Updates On every launch of a release build Keebye checks github.com/aiescu/keebye-releases for a newer version. If one exists it downloads, installs silently, and restarts the app. There is no prompt and no menu item for this. The update feed currently carries the Mac build only, so on Windows you update by downloading the newer installer from keebye.com. ## Where models are stored Models live under ~/Library/Application Support/com.keebye.app/models/ on a Mac and %APPDATA%\com.keebye.app\models\ on Windows. Downloads are resumable, verified with SHA-256, and retried up to five times. Only the default model downloads automatically; other engines download when you press "Download model" in Settings. - parakeet-v3 — about 670 MB (Settings shows ~640 MB). The default. - parakeet-v3-fp32 — about 2.55 GB (Settings shows ~2.4 GB). - canary-v2 — about 1.03 GB (Settings shows ~1 GB). - llm/Qwen_Qwen3-1.7B-Q4_K_M.gguf — about 1.28 GB, opt-in, used only for local-LLM polish.