Active Window Detection

What Speakly reads from the window you are in, and how to switch it off

To pick the right writing style and the right custom prompt, Speakly asks the operating system which window is in front when you dictate. The answer includes that window's title — often a document name, an email subject or the name of the person you are chatting with — so it is worth knowing exactly what is read, what it is used for and how to stop it.

What is read

  • The application name, for example Slack or Mail
  • The window title
  • The text visible in that window — but only when Agent Mode runs or a rule has Read my screen for context ticked. It is read through the accessibility tree, never as a screenshot, and goes only to your own AI model
  • The page URL, on macOS and only when the front app is a browser
  • The process id, and on macOS the bundle identifier

What it is used for

  • Choosing the writing style for post-processing: Speakly sorts the app into personal, work, email or other and applies the matching preference from the Style tab of the main window
  • Selecting a custom prompt you attached to that specific app

What the Permissions screen claims

Settings → Permissions marks the Screen Recording Access card as Optional and describes it like this: “Detects which app you're using, so Speakly can apply the writing style and the app-specific rule you set for it. We don't record your screen — no screenshots, no video: this permission only gives Speakly the name and title of the window in front.”

Your Dictionary is not modified
Nothing from a window title is ever saved to your Dictionary. The title is used, on your device, to pick a writing style and an app-specific prompt — nothing else. Earlier builds claimed otherwise on this very screen; that wording was corrected in v1.3.4.

The title stays on your device. The app uses it locally to choose a style and a prompt. The hint sent with your audio to a transcription provider is the style example plus your Dictionary words, and the window title is not part of it. The app name and the first 50 characters of the title are written to the local debug log. The *visible text* of the window is a separate thing: it is read only when Agent Mode runs or a rule asks for it, and it goes to your post-processing model — see Agent Mode.

Switching it off

  1. On macOS, detection depends on the Screen Recording permission. Open Settings → Permissions and use Open System Settings on the Screen Recording Access card
  2. Turn Speakly off under Privacy & Security > Screen Recording
  3. Back in Speakly, click Recheck Permissions to refresh the status

With the permission denied, detection returns nothing: Speakly falls back to your default style, and no custom prompt is applied at all — a rule only fires when it names the app you are dictating into, so with no app known, none matches. You can still pick a prompt by hand from the dot on the recording pill. On Windows there is no equivalent switch — window titles are readable there without any permission, so app detection cannot be turned off.

Active Window Detection — Speakly