Private voice typing means your spoken audio is never collected, stored, or routed through the app maker's servers, and ideally never leaves your PC at all. PipeVoice is a free, open-source Windows dictation tool with no account, no telemetry, and no servers of ours: in its fully offline mode (Local Whisper plus Ollama), nothing you say or type ever leaves your machine.
That is a stronger claim than most dictation apps can make, and because PipeVoice is open source you do not have to take our word for it. This guide explains what "private" should actually mean, shows exactly where your audio goes in each mode, and walks through the zero-data-leaves-your-PC configuration.
What "private" should mean for a voice typing app (and what it usually doesn't)
Most "private" claims in dictation software are thinner than they sound. A tool can call itself private while still requiring an account, phoning home with usage analytics, or sending every word you dictate to a cloud service for processing. Those are three separate concerns, and a genuinely private app has to address all of them.
- Account and identity: does it tie your dictation to a login, an email, or a device fingerprint?
- Telemetry and analytics: does it report what you do, how often, or what you say back to the vendor?
- Audio handling: where does the raw recording of your voice actually get processed, and who can read it?
A cloud-only tool can be perfectly reputable and still fail the third test by design, because it has to send your audio somewhere to turn it into text. The question is not whether a vendor is trustworthy, but how much you are forced to trust them.
Where your audio goes in each mode: local, BYO-key cloud, never our servers
PipeVoice lets you pick the transcription engine, and that choice decides where your audio travels. There is no hidden middle layer: PipeVoice never proxies your audio through any server we run.
| Engine | Where audio goes | Whose key | Cost |
|---|---|---|---|
| Local Whisper / faster-whisper | Stays on your PC, nothing sent | No key needed | Free |
| Deepgram (streaming) | To Deepgram only, on your key | Your free Deepgram key | ~pennies/day |
| OpenAI Whisper (batch) | To OpenAI only, on your key | Your OpenAI key | Pay-as-you-go |
If you choose a cloud engine, your audio goes directly from your machine to that one provider, using an API key you created and control. PipeVoice does not see it, store it, or relay it. If you choose Local Whisper, transcription runs entirely on your own CPU. The first time you use it, PipeVoice downloads a model of roughly 150MB, and after that no audio is transmitted anywhere. You can raise the model size for more accuracy at the cost of speed.
No account, no telemetry, no analytics on your voice: how PipeVoice is built
PipeVoice has no sign-up screen. You download the installer, run it, and start dictating. There is no email to hand over, no licence server to check in with, and no account that ties your dictation history to an identity.
There is also no telemetry. PipeVoice does not collect usage analytics, does not count your sessions, and does not report what you dictate. Your dictation history is stored locally on your own machine, not in a cloud account. The only network traffic PipeVoice generates on its own is the silent auto-updater, which fetches new versions and verifies them with a SHA-256 checksum before installing.
To dictate, you hold a hotkey (default Ctrl + \, or Right Ctrl), speak, and release. PipeVoice types real keystrokes into whatever app is focused: a terminal, an editor, a browser, or a chat box. A second hotkey copies the result to your clipboard instead of typing it. None of that requires a login.
Flow mode polish sends text only, never audio
PipeVoice has an optional AI cleanup step called Flow mode that removes filler words and fixes punctuation and casing. It is important to be precise about what this does to your data: Flow mode sends text only, never audio. The transcription happens first, and only the resulting text is passed to the polish provider.
You choose the polish provider too: OpenAI, Google Gemini (free tier), OpenRouter (free community models), or local Ollama. If you pick Ollama, the polish runs offline on your PC and no text leaves your machine either. If you pick a cloud provider, only the cleaned-up text transcript goes to it, on your own key.
The fully offline path: nothing leaves your PC
If your priority is that absolutely nothing leaves your machine, combine the two local options:
- Local Whisper / faster-whisper for transcription (offline, free, no key)
- Local Ollama for Flow mode polish (offline, no key)
This is the zero-cost, zero-key, zero-data-leaves-your-PC configuration. Your audio is transcribed locally, your text is polished locally, and PipeVoice itself sends nothing about your usage anywhere. It is the strongest privacy posture the app offers, and it is free.
Why open source lets you verify the privacy claims yourself
Privacy promises are only as good as your ability to check them. PipeVoice is open source, and the full code lives at github.com/Powleads/PipeVoice. You can read exactly where audio is sent, confirm there is no analytics SDK reporting your activity, and see that the only outbound calls are to the engine you selected and the update check.
If you do not want to read source, you can still verify the behaviour directly. Run the offline path, then watch your network activity with a tool like Windows Resource Monitor or a firewall: with Local Whisper and Ollama selected, you will see PipeVoice making no transcription traffic at all. That kind of independent verification is simply not possible with a closed, cloud-only app.
How cloud dictation tools differ on data handling
It helps to compare PipeVoice with common alternatives on the privacy axis specifically. None of this means the others are bad tools, only that their data model is different.
| Tool | Offline option | Account required | Open source | Notes |
|---|---|---|---|---|
| PipeVoice | Yes (Local Whisper + Ollama) | No | Yes | You choose the engine and bring your own key |
| Wispr Flow | No | Yes | No | Cloud-based, subscription |
| Google Docs Voice Typing | No | Yes (Google) | No | Cloud, tied to the Google Docs editor |
| Windows Voice Access | On-device | No | No | Built-in and free, but no engine choice |
| Otter.ai / Fireflies | No | Yes | No | Built around meeting transcription rather than typing into apps |
PipeVoice's wedge is the combination: free, open source, Windows-native, with an offline option, and it types into any focused app including the terminal and editors like Claude Code, Cursor, and VS Code. Read more in our Wispr Flow alternative comparison and on Windows alternatives to Wispr Flow.
Setting up the zero-data-leaves-your-PC configuration
- Download the installer from the latest release and run it. Because PipeVoice is currently unsigned, Windows SmartScreen may show an "unrecognised app" warning. Click More info, then Run anyway (code signing is in progress).
- In settings, choose Local Whisper / faster-whisper as your transcription engine. The first use downloads a model of about 150MB; raise the model size later if you want more accuracy.
- For Flow mode, choose local Ollama as the polish provider so cleanup also runs offline.
- Hold Ctrl + \ (or Right Ctrl), speak, and release. Your words are transcribed locally and typed into the focused app.
That is the whole setup. No key, no account, no recurring cost. For deeper configuration including per-app profiles, voice commands, and vocabulary boosting, see the docs. If you prefer a broader walkthrough first, our guides on offline voice typing for Windows and free voice typing software for Windows cover the wider picture.
Privacy checklist before choosing any dictation tool
Whatever app you end up using, run it through these questions:
- Does it require an account to dictate? (PipeVoice: no.)
- Does it send usage analytics or telemetry? (PipeVoice: no.)
- Is there an offline mode where audio never leaves your PC? (PipeVoice: yes, Local Whisper plus Ollama.)
- When using a cloud engine, do you control the key and the provider? (PipeVoice: yes, bring your own key.)
- Can you verify the claims, ideally because the code is open source? (PipeVoice: yes, on GitHub.)
- Does the AI cleanup step send your audio anywhere? (PipeVoice: no, text only.)
If a tool fails several of these, "private" is doing a lot of marketing work. PipeVoice was built to pass all of them, and to let you confirm it yourself.