Installation

Install SayIt

Pick your platform below — the guide for your system opens first. In short: download the installer, run it, follow the setup wizard, dictate. There is no account and no API key.

Requirements

Windows 10 or later, 64-bit
A working microphone
Network access for the first model download (roughly 100 MB to 1 GB depending on the model)

Download

Get SayIt-Setup-x64.exe from the download center or the GitHub Releases page. Verify the SHA-256 checksum shown with the release before running it.

Install

Run the installer. It installs only for your user (no administrator prompt needed), into %LOCALAPPDATA%\Programs\SayIt.
Optional tasks: a desktop shortcut and “Start SayIt when Windows starts” — both are off by default. A Start Menu entry is always created.
Leave Launch SayIt checked on the final screen (or start it from the Start Menu later).

About the SmartScreen warning

The installer is not code-signed yet. If Windows shows “Windows protected your PC”, choose More info → Run anyway — after confirming you downloaded it from the official release and checked the checksum.

Automated installs

The installer is a standard Inno Setup package and accepts the usual silent flags (/SILENT, /VERYSILENT) for unattended deployment.

First launch & verification

SayIt runs in the system tray. On first start, the setup wizard opens — that window is the verification that the install worked. SayIt has no command-line version flag; success looks like: tray icon visible, wizard open, and a first dictation inserting text (after the model is installed). Continue with the first-run setup below.

Troubleshooting

Windows SmartScreen shows “Windows protected your PC”

The installer is not code-signed yet, so Windows shows this warning for new downloads. Choose More info → Run anyway — but only because you downloaded the installer from the official releases page. The site does not claim any signing verification, and you can check the SHA-256 checksum against the value shown on the download page.

The microphone is not detected

Check the input device selection in the setup wizard or in Settings → Configuration, and verify Windows microphone permissions for SayIt. Use the built-in microphone test to confirm a signal.

The model download fails or is slow

The first model download needs network access and can be roughly 100 MB to 1 GB depending on the model. It is stored under %LOCALAPPDATA% \SayIt\models and is only downloaded once per model.

The hotkey does nothing in some apps

Some applications intercept or block global shortcuts. Try a different combination in Settings → Configuration — no single hotkey works in every application on Windows.

Text is not inserted after dictation

Insertion requires a focused editable text field in the active app. If a paste fails, the transcript stays recoverable — the app shows an error state instead of silently losing the text, and the last transcript remains on the clipboard.

SayIt seems stuck on recording or processing

Press Esc to cancel the active dictation. If the tray icon stays unresponsive, quit from the tray menu and relaunch.

Requirements

A 64-bit Linux distribution with glibc 2.28 or newer (e.g. Ubuntu 20.04+) — the AppImage is built against manylinux_2_28
xclip for clipboard-based text insertion
A working microphone; network access for the first model download
X11 session for global hotkeys (Wayland capture is a known limitation)

Download

Get the SayIt-*.AppImage from the download center or the GitHub Releases page, and verify its SHA-256 checksum.

Make it executable and run

1
Make the AppImage executable:
Linux shell
```
chmod +x SayIt-*.AppImage
```
2
Start SayIt:
Linux shell
```
./SayIt-*.AppImage
```
Optional: keep the AppImage somewhere stable such as ~/Applications. SayIt does not install desktop integration automatically; it lives where you put it.

First launch & verification

SayIt runs in the system tray, and the setup wizard opens on first start — seeing that wizard verifies the AppImage works. Note the honest limits: global hotkeys may require X11, and clipboard insertion needs xclip. Continue with the first-run setup below.

Troubleshooting

Running the AppImage says “Permission denied”

The executable bit is not set (or was lost when the file was moved by an unzipping tool). Run chmod +x SayIt-*.AppImage again, then start it with ./SayIt-*.AppImage.

Text is not inserted (clipboard issues)

Clipboard operations require xclip. Install it with your distribution’s package manager (for example sudo apt install xclip on Debian/Ubuntu).

The global hotkey does not respond

Global hotkey capture may require X11 and may not work under Wayland. This is a known limitation of the current Linux support, not a misconfiguration on your side.

The model download fails

Model downloads need network access; weights are fetched from the upstream Sherpa-ONNX releases and stored locally. Retry from the setup wizard or Settings once connectivity is restored.

Developer / source install

This route is for development and contribution. Regular users don’t need any of it — use the Windows installer or the Linux AppImage.

Requirements

Python 3.12 or newer
The uv package manager
Git
Network access for the first model download

Clone and sync

Clone the repository:

Shell

git clone https://github.com/VptrCipher/SayIt.git

2
Create the virtual environment and install dependencies:
Shell
```
cd SayIt && uv sync
```

Run

Shell

uv run python -m sayit

Why python -m?

The sayit entry points declared in pyproject.toml are not installed into the environment by uv sync — run the app as a module.

Run the tests

Windows PowerShell

$env:QT_QPA_PLATFORM = "offscreen"
$env:PYNPUT_BACKEND = "dummy"
uv run pytest -m "not slow"

The same suite on Linux/macOS:

bash

QT_QPA_PLATFORM=offscreen PYNPUT_BACKEND=dummy uv run pytest -m "not slow"

Verification

uv run python -m sayit starts the tray app and opens the setup wizard on first run — that is the check. A passing fast test suite verifies the code, not the packaged app.

First-run setup

On first launch, the setup wizard walks you through everything. It states honestly what is and isn’t ready — for example, it never fakes a transcription while no model is installed.

Welcome & privacy — what SayIt does and a plain-language privacy summary.
Microphone — select your input device and run the built-in signal test.
Speech model — install a model now or finish setup and do it later. Model downloads are separate from the application and are roughly 100 MB to 1 GB depending on the model.
Hotkey — choose the push-to-talk combination. The default is CtrlSpace.
Try your first dictation — hold the hotkey, speak, release; the text appears at your cursor. Esc cancels mid-dictation.
Setup complete — a summary of what is ready and what remains (dictation stays unavailable until a model is installed).

One honest note

You can complete the wizard without a model, but you can’t dictate until one is installed — the app tells you so instead of pretending.

Uninstalling — and what happens to your data

Windows

Uninstall from Settings → Apps → Installed apps → SayIt (a running instance is closed automatically). Your user data — settings, vocabulary, history, logs, and downloaded models in %LOCALAPPDATA%\SayIt — is kept by default; the uninstaller asks before deleting it. For a full wipe at any time, use Clear Data in Settings → Privacy & Data.

Linux

Delete the AppImage file — that removes the application. Your settings and downloaded models are stored separately in your user directories and are not removed automatically; run Clear Data in the app first if you want them gone.