CLASSIFICATION: PUBLIC DOC-ID: JGS-VOICEFLOW · DOCS REV 0.2.0

§ Docs

VoiceFlow documentation

Quickstart, FAQ, troubleshooting, the privacy guarantee and how to verify it, and how to check your download. For the full walkthrough see the guide.

Quickstart

From download to your first dictation in a few minutes. The one-time download and first-run setup take a little longer (see the notes below), but the hands-on steps are quick.

1. Download

Go to the latest release and download VoiceFlow-Setup.exe.

Heads up: the installer is about 700 MB because the offline speech models are bundled inside it, so you can dictate with no internet afterward. On a typical connection the download takes a few minutes. If you like, verify the download before running it.

2. Run the installer

Double-click VoiceFlow-Setup.exe.

Windows SmartScreen / "Windows protected your PC". VoiceFlow is a free release and is not code-signed, so Windows shows a blue "Windows protected your PC, unknown publisher" screen. This is expected. Click More info, then Run anyway.

The installer is per-user (no administrator rights needed). It adds a Start-Menu shortcut and offers an optional start-on-login checkbox.

3. First launch (one-time auto-configuration)

The first time VoiceFlow runs, it configures itself for your hardware: it checks your CPU, memory, and GPU and picks the best local speech model and settings.

4. Your first dictation

  1. Place your cursor where you want text (any app: editor, browser, chat).
  2. Hold Ctrl+Space, speak a sentence, then release.
  3. VoiceFlow transcribes locally, cleans the text, and pastes it at your cursor.
The very first transcription after launching (or after an update) also loads the speech model into memory, which adds a one-off few-second delay. After that, dictation is fast.

5. Settings and modes

Open the tray icon, then Settings, to change the hotkey, output mode, and providers. The output mode (raw / clean / summary / prompt / code) can also be switched per capture from the tray menu. See the guide for what each mode does and how to enable optional LLM cleanup. Settings live in %APPDATA%\VoiceFlow\.

FAQ

Is it really free?

Yes. VoiceFlow is free to use. It is proprietary software (the source is not public, and you may not copy, modify, or redistribute it), but you can install and run it free of charge for personal or internal business use. See the LICENSE.

Does it work offline?

Yes. On the default settings, speech-to-text and cleanup both run on your machine, and the installer bundles the speech models, so you can dictate with no internet at all.

Do I need a GPU?

No. VoiceFlow runs on CPU. An NVIDIA GPU (with CUDA) makes transcription faster, but it is optional. AMD and Intel GPUs are not used for acceleration; those machines fall back to CPU, which is still fast enough for everyday dictation.

Which languages does it support?

VoiceFlow uses Whisper, which supports many languages. It works best when you set your language explicitly in Settings (that is also faster than auto-detect).

How do I get better cleanup (summary and prompt modes)?

The clean mode works offline with no setup. For summary and prompt, results are better with a language model:

Both are opt-in and configured in Settings or the config file. If no LLM is configured, summary and prompt fall back to the offline cleaner.

Does it send my voice anywhere?

No, not on the default path. Audio and text stay on your machine. Cloud providers are opt-in only. See privacy for the full statement and a recipe to verify it yourself.

How do I uninstall VoiceFlow?

Uninstall it like any Windows app: Settings, Apps, Installed apps, VoiceFlow, Uninstall (or use the "Uninstall VoiceFlow" Start-Menu shortcut). That removes the per-user install, including the bundled speech models. A couple of things are left behind on purpose so you do not lose data if you reinstall (remove them manually for a clean wipe): your settings and dictation history in %APPDATA%\VoiceFlow\, and any extra speech models you downloaded later (in the faster-whisper / Hugging Face model cache).

How do I update to a new version?

VoiceFlow v0.1.x has no auto-update. To update, download the latest VoiceFlow-Setup.exe from the releases page and run it over your existing install. Your settings and history in %APPDATA%\VoiceFlow\ are preserved. Each installer is the full download; there is no smaller delta update yet.

Is the source code available?

No. VoiceFlow is proprietary. It is not code-signed yet, so download only from the official releases page and verify the SHA-256 checksum to confirm your copy is intact.

Troubleshooting

Antivirus or "Windows protected your PC" (SmartScreen) warnings

This is the most common first-run question. VoiceFlow's installer is a native, compressed Windows binary, and it is a free release that is not code-signed (a code-signing certificate is a paid, recurring cost). Two things can trigger a warning:

  1. SmartScreen "unknown publisher". Because the installer is not signed, Microsoft SmartScreen cannot identify a publisher, so it shows "Windows protected your PC" with an unknown-publisher message. This is expected for an unsigned app; it is not a sign that anything is wrong with the file.
  2. Antivirus heuristics. Some antivirus engines flag freshly released, compressed, unsigned executables as a precaution (a false positive), even when nothing is wrong.

What to do: on the SmartScreen screen, click More info, Run anyway. If your antivirus quarantines the installer, you can allow or restore it in your AV's quarantine or exclusions screen. For peace of mind, download only from the official releases page and verify the SHA-256 checksum before running it.

No microphone / wrong device

If VoiceFlow cannot find a microphone, you get a clear error on the first capture (it does not fail silently). Check Windows Settings, System, Sound, Input and make sure the right device is set as default and is not muted, then try again.

My GPU is not being used

VoiceFlow only accelerates transcription on NVIDIA GPUs (CUDA). AMD and Intel GPUs are not used for acceleration; those machines run on CPU, which is still fast enough for everyday dictation.

The first dictation is slow or looks frozen

The first transcription after you launch VoiceFlow (or after an update) loads the speech model into memory, which adds a one-off delay of a few seconds. This is normal; subsequent dictations are fast. If it seems stuck for much longer than that, check the microphone advice above and the logs.

Models and offline

The installer bundles the speech models, so transcription works with no download. If you select a model that was not bundled, VoiceFlow may need to fetch it the first time, which requires a network connection.

Hotkey conflicts

If Ctrl+Space (or another hotkey) clashes with another app, rebind it in the tray, Settings, Hotkeys.

Where are the logs and settings?

In %APPDATA%\VoiceFlow\: paste that into the File Explorer address bar. You find the config and log files there, handy to attach to a bug report.

Privacy

On the default path, nothing you say or type with VoiceFlow leaves your machine. This section explains exactly what that means, and how you can verify it yourself rather than take our word for it.

The guarantee

With the default settings (or a fresh install with no changes):

What is opt-in (and off by default)

VoiceFlow can use cloud services, but only if you turn them on. Cloud speech-to-text and cloud LLM cleanup are disabled by default. Enabling them requires editing the configuration to point at a provider and supply credentials, so it cannot happen by accident. If you enable a cloud provider, then (and only then) the relevant audio or text is sent to that provider so it can do the work. That is the trade-off you opt into for cloud quality and speed.

One caveat worth knowing. If you configure a local LLM through a tool like Ollama, make sure you use an on-device model. Some model names that look local (for example, anything ending in :cloud) actually route through a remote server, so your text would leave the machine despite the "local" label. For true zero-egress, use a genuinely on-device model.

Verify it yourself

You do not have to trust us. You can prove that a default dictation sends nothing over the network:

  1. Block VoiceFlow from the network. Either create an outbound block rule in Windows Defender Firewall for the installed app at its full path (%LOCALAPPDATA%\Programs\VoiceFlow\VoiceFlow.exe; a bare program name will not create a working rule, so use the full path), or simply disable your network adapter / unplug for the test.
  2. Dictate as normal: hold the hotkey, speak, release.
  3. Confirm the text still appears at your cursor, with no network traffic (watch Resource Monitor, Network, or note that the firewall rule did not prompt).

If transcription and cleanup still work with the network blocked, you have confirmed the default path is fully local.

This recipe verifies the running app's network use, not the installer download. For that, see verify your download. It also assumes the standard installer; a portable build would run from a different path, in which case disabling the network adapter is the build-independent check.

Verifying your download

Every VoiceFlow release ships with a SHA-256 checksum so you can confirm your VoiceFlow-Setup.exe downloaded correctly and has not been tampered with before you run it. The check takes under a minute in PowerShell.

VoiceFlow is not code-signed yet. This is a free release and does not carry a paid code-signing certificate, so Windows shows an "unknown publisher" warning when you run the installer (see troubleshooting), and you cannot verify a publisher signature. The SHA-256 checksum is therefore the way to confirm your download is intact and matches the official release. Always download from the official releases page, never from a mirror or third-party site.

Check the SHA-256 checksum

Each release includes a SHA256SUMS.txt file alongside the installer. Download both from the official release, then in PowerShell (in the folder where you saved them):

Get-FileHash .\VoiceFlow-Setup.exe -Algorithm SHA256
Get-Content .\SHA256SUMS.txt

The hash printed by Get-FileHash must match the one in SHA256SUMS.txt (case-insensitive).

If the checksum does not match

If the hashes differ, do not run the installer: your download is corrupted or has been tampered with. Re-download from the official releases page. If it still does not match, please report it privately via SECURITY.

Because the checksum file is published on the same release page as the installer, it confirms the file is intact and matches what we published. It does not, on its own, prove who built it (that is what a code signature would add). Downloading only from the official release page is what ties the file to us.

Read the full guide, back to the home page, or download VoiceFlow.