Skip to main content

Advanced Settings

Everything here is optional. You can run Knowii Voice AI for months without opening this page. Come back when you want the app to start with your session, when the overlay lands in the wrong place, when pasting misbehaves on Linux, or when you need the logs.

Application Behavior​

Control how Knowii Voice AI starts and runs:

Start Hidden​

Location: Settings > Advanced

  • Launch directly to system tray without showing the main window
  • Useful if you want the app running silently in the background
  • The app will be fully functional and ready to use via the system tray icon

Show Tray Icon​

Location: Settings > Advanced

  • Show or hide the icon in your system tray (near the clock). On by default
  • Turning it off removes the icon immediately — no restart needed — and turning it back on restores it
  • With the tray icon hidden, the app protects you from losing the window: closing it minimizes instead of hiding it, "Start Hidden" is ignored at the next launch (the window always appears), and you can quit from the button at the bottom of the main window
  • Useful for minimal setups — for example a status bar that triggers dictation through the CLI instead of the tray

Minimize to Tray​

Location: Settings > Advanced

  • Off by default. When on, minimizing the window tucks it away into the system tray instead of leaving it in your taskbar or dock
  • Bring it back the usual way: click the tray icon, or right-click it and choose Settings...
  • Closing the window already sends the app to the tray — this setting is for people who want the minimize button to do the same
  • It only applies when there is a tray icon to bring the window back from, so it does nothing while Show Tray Icon is off, when the app was started with --no-tray, or on a GNOME session without a tray extension. In those cases minimizing behaves normally

Launch on Startup​

Location: Settings > Advanced

  • Automatically start when you log in to your computer
  • App will be ready to use immediately when you need it
  • Combined with "Start Hidden", the app starts with your session and stays out of sight

Check for Updates Automatically​

Location: Settings > Advanced

  • When on (the default), the app looks for a new version once at startup
  • Turn it off and the app makes no automatic network requests at all — checking for updates is its only recurring one
  • You can always check manually from the system tray menu, the status bar, or Settings > About, and update by hand whenever you like

Display Options​

Appearance​

Location: Settings > Advanced > Display

Choose how Knowii Voice AI looks:

  • Dark (default): the classic look the app has always had
  • Light: a bright variant of the interface
  • System: follows your operating system's theme, switching automatically when it changes

The choice applies immediately to the app window and the recording overlay.

Overlay Enabled​

Location: Settings > Advanced > Display

Enable or disable the visual overlay that provides feedback during recording and transcription:

  • Enabled (default): Shows the overlay during recording and transcription — and, when AI post-processing is on, while the cleanup pass runs (🤖 Post-processing using AI)
  • Disabled: Hides the overlay completely
    • Useful during screen recordings, presentations, or screen sharing
    • No visual indication will appear during transcription

The overlay displays:

  • Recording indicator when you're actively recording
  • Transcription progress when processing
  • Cancel option to abort the current operation

Overlay Position​

Location: Settings > Advanced > Display

Only available when Overlay Enabled is turned on

Choose where the recording overlay appears on your screen:

Center Positions:

  • Bottom Center (default): Status shown at the bottom center of your screen
  • Top Center: Status shown at the top center of your screen

Corner Positions:

  • Top Left: Status shown in the top-left corner
  • Top Right: Status shown in the top-right corner
  • Bottom Left: Status shown in the bottom-left corner
  • Bottom Right: Status shown in the bottom-right corner

The overlay keeps clear of taskbars, menu bars, and panels, so it never lands on top of something else.

Transcription Options​

Translate to English​

Location: Settings > Advanced > Transcription

  • Automatically translate speech from other languages to English
  • Only available with Whisper models (not supported by Parakeet)
  • Useful for multilingual workflows where you speak in one language but need English output
  • Examples: Speaking French but need English documentation, translating meeting notes, etc.

Note: Translation requires more processing time than standard transcription.

Transcribe While Recording (faster stop)​

Location: Settings > Advanced > Transcription

Called "Instant transcription" until 0.9.0. Renamed because the old name promised text appearing while you speak — that is a different feature, Live dictation, just below.

  • Off by default. When on, Knowii Voice AI transcribes your speech at natural pauses while you are still talking, instead of waiting until you stop.
  • When you stop, only the last few words still need processing, so the wait after releasing the key is much shorter — even after a long dictation.
  • Nothing appears while you talk: the text still arrives in one piece, at the end, exactly like before. The recording saved to your history is also unchanged.
  • If AI post-processing is on, its cleanup pass still runs after the transcription — a few seconds with the command-line agents, about a second with a local or API model — so that is the wait you will notice, not the transcription.
  • Works with every model.
  • Trade-off: the audio is cut at pauses, so the model occasionally loses a bit of context across a pause. If you notice odd wording around pauses, turn it off — the default whole-recording mode is unchanged.

Tip: this shines on long dictations. A two-minute thought that used to take several seconds to transcribe is ready the moment you release the key.

Live Dictation​

Location: Settings > Advanced > Transcription

  • Off by default. When on, your words appear in the target window while you keep talking: each chunk is pasted the moment it is transcribed, at the natural pauses in your speech — typically well under a second after you pause, every few seconds during a long dictation.
  • It builds on Transcribe while recording, which is shown as on and locked while Live dictation is on.
  • Stopping still does the last bit: the words after your final pause are transcribed and pasted, the whole text is saved to History as one entry, and the trailing space, clipboard copy and auto-submit options are applied once, at the end — not after every chunk.
  • Cancel stops the stream, it does not erase it. Pressing the cancel shortcut mid-sentence stops further chunks from arriving; what is already on screen stays, because the app cannot know what the other program did with the text. As with any cancelled dictation, nothing is saved to History.
  • Turns AI post-processing off. The cleanup pass needs the whole text, and Live dictation has already pasted it piece by piece — the two cannot both be on. Switching Live dictation on turns the cleanup pass off (with a notice), and its switch stays greyed out with the reason until you turn Live dictation off again. Your custom words, word replacements and transcription hook still apply — to each chunk.
  • Works with every model and every paste method. With a clipboard-based method (Ctrl+V, Shift+Insert, Ctrl+Shift+V) the app saves and restores your clipboard around every chunk, as it does for a normal paste; the Direct method types the text and leaves the clipboard alone, so it is the smoothest choice. With Clipboard only, nothing is pasted while you talk — the whole transcript lands on the clipboard at the end, as usual.
  • Keep the target window focused while you dictate: each chunk goes to whatever window is active when it arrives, exactly like a normal paste. Switching windows mid-dictation sends the rest of your words there.
  • Same trade-off as Transcribe while recording: the audio is cut at pauses, so the model occasionally loses a little context across a pause.

Tip: pair it with a headset and a long-form editor. Notes, drafts and messages fill in as you think aloud, and the wait at the end is the last few words only.

Transcription Hook (Advanced)​

Location: Settings > Advanced > Transcription

  • Off by default. When on, every transcription is handed to a small program you write before it is pasted — your program receives the text and whatever it prints back is what gets pasted (and saved to History).
  • This is a power-user feature in the spirit of git hooks. A few things people use it for:
    • Clean up the text your way: strip trailing periods, lowercase the first letter, expand personal shorthand
    • Pipe the text through a local AI (like Ollama) to rewrite, translate, or summarize it
    • Send the text somewhere else — open a link with it, call a webhook, append it to a file
  • Setting it up: turn the toggle on, click Open Hooks Folder, and create a script named transcription in that folder (on Windows: transcription.bat, transcription.cmd, or transcription.exe). On Linux and macOS, make it executable (chmod +x transcription). The script reads the text from standard input and prints the replacement text to standard output.
  • Your words are safe: if the script is missing, crashes, or takes longer than 30 seconds, the original transcription is pasted unchanged. If the app window is open you'll see an error explaining what went wrong; otherwise the reason is recorded in Settings > Advanced > Application Logs.
  • If your script prints nothing (and exits successfully), nothing is pasted and nothing is saved — useful when the script delivers the text somewhere itself.

Example script (Linux/macOS) that removes a trailing period:

#!/bin/sh
sed 's/\.$//'

Privacy note: the hook runs entirely on your computer. Knowii Voice AI never sends your text anywhere — but your own script can, so only use scripts you trust.

What the script may look like (Linux and macOS): because the app runs it with every dictation, the script and the hooks folder must belong to you (or to root) and must not be writable by other users. A script that fails this check is refused, your text is pasted unchanged, and the error tells you what to change — usually chmod go-w on the file or the folder. The script starts in the hooks folder, so relative paths in it resolve there, and it gets your normal environment rather than the app bundle's private library paths — so python, curl or jq behave exactly as they do in your terminal. The path of the script in use is written to the log once.

AI Post-Processing​

:::info Coming in the next release

AI post-processing is not in version 0.9.0. It is described here ahead of the release it ships in — check Settings > About for your version.

:::

Location: Settings > Advanced > AI Post-Processing

  • Off by default. When on, every transcription is handed to an AI model that fixes punctuation, capitalization and obvious mistakes before it is pasted — turning "so um this is a a test" into "This is a test."
  • Not available while Live dictation is on: the cleanup needs the whole text, and Live dictation pastes it as you speak. The switch is greyed out with the reason; turn Live dictation off to use it.
  • Ollama is the default provider: it runs on your own computer, needs no account, and sends nothing anywhere. OpenAI, OpenRouter, the Claude API, and the claude / codex command-line tools are also supported, as is any OpenAI-compatible server you point it at.
  • A provider that sends text off your computer is blocked behind an explicit confirmation, asked per provider and revocable at any time. Your audio recordings are never sent, by any provider.
  • Your words are never lost: if the service is unreachable, too slow, or returns something odd, your original transcription is pasted unchanged and a notification tells you why.
  • Give up after controls how long a cleanup may take before your original text is pasted instead (one minute by default).
  • A Test button runs a fixed sample sentence — never your own text — so you can check a setup, and see how long it takes, without dictating.

See the AI Post-Processing guide for setup, API keys, model choice, and troubleshooting.

Paste Options​

Paste Method​

Location: Settings > Advanced > Paste

Controls how transcribed text is inserted into applications:

  • Clipboard (Ctrl+V) (default on Windows/macOS): Simulates Ctrl/Cmd+V keystrokes to paste

    • Most compatible with applications
    • Works in virtually all text fields
    • Temporarily modifies clipboard (see Clipboard Handling below for behavior options)

  • Clipboard (Shift+Insert) (Windows/Linux only): Simulates Shift+Insert keystrokes to paste

    • More universal for terminal applications
    • Works better in some Linux/X11 applications
    • Alternative to Ctrl+V for applications that don't respond well to it
    • Temporarily modifies clipboard (see Clipboard Handling below for behavior options)

  • Clipboard (Ctrl+Shift+V) (Windows/Linux only): Simulates Ctrl+Shift+V keystrokes to paste

    • The paste shortcut of most terminal emulators (GNOME Terminal, Konsole, Alacritty, Windows Terminal, …), where plain Ctrl+V is taken
    • Layout-independent like Ctrl+V, so it is never overridden by the app on desktops where direct typing is unsafe
    • Temporarily modifies clipboard (see Clipboard Handling below for behavior options)

  • Direct (default on Linux): Types the text straight into the focused field

    • Bypasses clipboard completely
    • On Linux, uses your desktop's typing tool (wtype/kwtype/xdotool) for reliable results. See Typing Tool below
    • Useful for troubleshooting clipboard-related issues

  • Clipboard only (no paste): Copies the transcription and stops there

    • Nothing is typed or pasted anywhere; the text waits on the clipboard and in History until you paste it yourself
    • Available on every desktop, including those where the app otherwise manages the paste method (GNOME/Wayland) — it is the one choice that is never overridden
    • Auto-submit does nothing in this mode, since nothing was pasted

When to change:

  • If transcriptions aren't being pasted correctly
  • If clipboard conflicts occur with other applications
  • For troubleshooting pasting issues in specific applications
  • If using terminal applications on Linux/Windows (try the Shift+Insert or Ctrl+Shift+V method)
  • If you would rather decide where each dictation goes (Clipboard only)

Typing Tool (Linux only)​

Location: Settings > Advanced > Paste

Chooses which system tool Knowii Voice AI uses to type transcriptions and send paste shortcuts on Linux. This matters because typing text reliably differs between Wayland and X11.

  • Auto (default): Automatically picks the best tool available on your system: wtype (or kwtype on KDE) on Wayland, xdotool on X11.
  • wtype / kwtype / dotool / ydotool / xdotool: Force a specific tool.
    • dotool tip: if you run its daemon (dotoold), the app talks to it through dotoolc and skips creating a virtual keyboard on every dictation — noticeably faster to the first character. It looks for the daemon's pipe at $DOTOOL_PIPE, else $XDG_RUNTIME_DIR/dotool-pipe, on every use, so starting or stopping dotoold takes effect immediately.

The dropdown only lists the tools actually installed on your computer. If your list shows just "Auto", install one of the tools (see the Installation guide). Otherwise the app falls back to a built-in method that can be unreliable on Wayland (text may fail to appear).

When to change:

  • If dictated text doesn't appear in your apps on Linux
  • If you have multiple tools installed and want to force a particular one

Clipboard Handling​

Location: Settings > Advanced > Paste

Control what happens to your clipboard after transcription:

  • Don't Modify Clipboard (default): Preserves your current clipboard contents

    • Transcription is pasted but your clipboard remains unchanged
    • Useful if you have something important copied that you don't want to lose
    • This works for text and images (a screenshot you just took survives a dictation). On Wayland with wl-clipboard installed, copied files are put back too. (Images and files: coming in the next release; version 0.9.0 keeps text only.)
    • Only one form of what you copied comes back: an image is restored as an image, files as a file list, text as plain text. Extras some apps add (formatting, an app's private data) are not. When an app copies both text and an image (cells in a spreadsheet, for example), the text comes back
    • Something that cannot be read at all (an unusual format, or an app that does not answer in time) is left alone, and the transcription stays on the clipboard instead

  • Copy to Clipboard: Leaves transcription in clipboard after pasting

    • Allows you to paste the same transcription multiple times
    • Useful for repeated pasting or when you want to keep the text in clipboard

Paste Delay​

Shown for the clipboard-based paste methods only. It is the pause between putting your transcription on the clipboard and pressing the paste shortcut — the time the application you are dictating into gets to regain focus after the shortcut and the recording indicator.

  • Default: 80 ms, which is what the app always used before this became a setting
  • Range: 10 ms to 200 ms, as presets
  • Raise it if an application sometimes pastes nothing, or pastes the previous clipboard content, right after you dictate — the classic sign of a paste that arrived before the app was listening. Electron-based apps and remote desktops are the usual candidates.
  • Lower it if pasting feels sluggish in apps that are always ready

The app restores whatever was on your clipboard before the dictation once the paste has landed. That restore waits at least 400 ms, and longer when you raise the paste delay, so a slow application can never end up pasting your old clipboard instead of the transcription.

Trailing Space After Paste​

Location: Settings > Advanced > Paste

Default: Off

Append a space after each pasted transcription. Turn this on if you dictate in several takes and your snippets end up glued together — each new dictation then lands one space after the previous one.

Auto-Submit After Paste​

Location: Settings > Advanced > Paste

Default: Off

Press the submit key automatically right after your transcription is pasted — dictate a message and it sends itself. Made for chat apps and AI assistants: speak, release the shortcut, and the message goes out with no keyboard touch.

When enabled, a Submit Key dropdown appears:

  • Enter (default): Submits in most chat apps
  • Ctrl+Enter (Cmd+Enter on macOS): For apps where plain Enter adds a new line instead of sending
caution

With this on, every dictation is sent immediately — there is no moment to review or edit the text first. Keep it off for dictating into documents or emails.

Input Devices (Linux only)​

Location: Settings > Advanced

Disable Shortcut Suppression​

Default: Off

On Linux, Knowii Voice AI listens to your keyboard directly so your shortcut works in every app. While listening, it takes temporary exclusive control of the devices it recognises as keyboards, so your shortcut keys don't also land in whatever you're typing in.

On rare hardware (usually gaming mice and fancy keyboards, which present themselves as a keyboard and a mouse through the same connection) this can affect the wrong device.

Turn this on if, while the app is running:

  • the mouse pointer stops moving but clicks still work,
  • a key, media button, or scroll wheel stops responding, or
  • trackpad gestures stop working.

With it on, your shortcut still works and your devices are left completely alone. The trade-off is that the shortcut's keys also reach the app you're typing in, so you may see a stray character or hear a beep when you trigger it.

note

This setting takes effect the next time you start the app.

tip

Most people never need this. Devices that report pointer movement are already left alone automatically. See My mouse or keyboard behaves strangely while the app is running in the FAQ.

History Management​

Location: Settings > Advanced

Control how transcriptions are saved and managed:

Enable History​

  • Stores transcriptions in the database
  • Allows you to view, search, and manage past transcriptions
  • Access history from the History tab in the main window
  • When disabled, transcriptions are not saved anywhere

Save Transcriptions to Disk​

Requires History to be enabled

  • Saves transcription text as markdown files
  • Files organized by date in history/YYYY/MM/ folders
  • Example: 2025-01-15-14_30_45 - Transcription.md
  • Useful for backups or accessing transcriptions outside the app

Save Audio to Disk​

Requires History to be enabled

  • Saves audio recordings as WAV files
  • Allows playback of original recordings from history
  • Files stored alongside transcriptions: 2025-01-15-14_30_45 - Audio.wav
  • Warning: Audio files can consume significant disk space

Keep Recordings​

Requires History and "Save Audio to Disk" to be enabled

  • Automatically deletes audio recordings older than the chosen window: Forever (the default — nothing is ever deleted), 1 day, 1 week, 1 month, or 3 months
  • Only the audio files are removed — the transcription text and the history entry stay
  • Starred entries are never touched: star a transcription to keep its recording no matter what
  • A privacy and disk-space win: recordings are by far the largest files the app produces
  • Cleanup runs at startup, after each transcription, and immediately when you pick a shorter window

Open Recordings Folder​

Requires History and "Save Audio to Disk" to be enabled

  • Opens the folder where your audio recordings are stored in your file manager
  • Respects a custom audio folder if you configured one; otherwise opens the default history folder

Limit History​

Requires History to be enabled

  • Keeps only the most recent unstarred transcriptions
  • Default limit: 100 most recent entries
  • Starred transcriptions are always kept regardless of this setting
  • Helps manage disk space automatically
  • Older unstarred transcriptions are automatically deleted

Clear All History​

Requires History to be enabled

Warning: This action is permanent and cannot be undone!

  • Permanently deletes all transcriptions (including starred entries)
  • Deletes all associated files (text and audio)
  • Clears the database completely
  • Use with caution, and consider backing up your data first

Performance Settings​

Model Unload Timeout​

Location: Settings > Advanced > Performance

Control when the AI model is automatically unloaded from memory to free up system resources:

  • Never: Model stays loaded in memory (fastest, uses more RAM/VRAM)
  • Immediately: Unload right after each transcription (slowest, minimal memory usage)
  • After 2/5/10/15 minutes: Balanced options
  • After 1 hour: For regular but not constant use

Recommendations:

  • Never: If you use transcription frequently throughout the day and have sufficient RAM
  • After 2-5 minutes: Balanced option for moderate use
  • After 10-15 minutes: Good for occasional use
  • Immediately: Only if you're very constrained on memory

When a model is unloaded, the next transcription will have a delay while the model reloads.

info

The AI model now stays ready throughout a recording. Even if you've chosen to unload the model quickly, it won't be unloaded in the middle of a recording, so your transcription is never interrupted while you're speaking.

Use GPU for Whisper​

Location: Settings > Advanced > Performance

When you use a Whisper model, Knowii Voice AI normally runs it on your graphics card for faster transcription. If your graphics card or its drivers cause trouble — crashes, garbled text, or transcriptions that never finish — turn this off to run Whisper on your processor instead. Transcription becomes slower but very reliable.

The change applies the next time the model loads. To apply it right away, switch to another model and back, or restart the application.

tip

Leave this on unless you are experiencing problems. On machines without a compatible graphics card, Knowii Voice AI already falls back to the processor automatically.

Application Logs​

Location: Settings > Advanced > Application Logs

View the most recent application activity directly inside the app, with no need to find log files on your computer:

  • Refresh: Loads the latest log lines
  • Copy: Copies the logs to your clipboard so you can paste them into an email or a bug report

This is the fastest way to gather information when reporting a problem. The logs never contain the words you dictated. They only describe what the app is doing.

Tips​

  • Start with default settings and adjust based on your needs
  • Monitor disk space if saving audio files
  • Use "Limit History" to prevent excessive disk usage
  • Combine "Start Hidden" and "Launch on Startup" so the app is always ready and never in the way
  • Disable the overlay during screen recordings, presentations, or screen sharing
  • Use corner positions for the overlay if center positions interfere with your work

Need Help?​

If you have questions about advanced settings, visit the Support page.