"We don't collect your data" is the new "we take your privacy seriously." It's said so often it's stopped meaning anything. Most companies that say it are technically accurate in some narrow sense while still doing something with your data you wouldn't love if you thought about it.
We want to be more specific about what Tellaflow does, because the specifics matter, especially for a tool that processes your voice.
What a typical cloud voice tool does with your audio
When you use a cloud-based voice transcription tool, the flow looks like this:
- You speak into your microphone
- Your Mac records an audio buffer
- That buffer is sent, encrypted, to a server over the internet
- A model running on that server processes the audio
- Text is sent back to your Mac
- The audio buffer may or may not be retained, depending on the privacy policy
Even in the best-case scenario where the vendor deletes your audio immediately after transcription, several things are still true:
- Your audio existed on a server you don't control, however briefly
- It was visible to the vendor's infrastructure at the time of processing
- The text transcription may be retained longer than the audio
- API intermediaries (like OpenAI's API) may have their own retention policies
- Security breaches are always possible regardless of intentions
What Tellaflow does with your audio
Here is the complete list of places your audio goes when you use Tellaflow:
- Your microphone
- Your Mac's RAM (where the audio buffer lives for processing)
- The Whisper or Parakeet model running in your Mac's Neural Engine
- The text output, which goes wherever your cursor is
That's it. Your audio never leaves step 4. There is no step 5 where it goes somewhere else.
The reason we can say this confidently more confidently than a company with a server is that there is no server. Tellaflow doesn't have a backend that receives audio. There's no endpoint to send it to even if we wanted to.
The open-source verification
We're not asking you to trust us. We're open source.
You can read every line of Tellaflow's code at github.com/vv-zero-oss/tellaflow-app. You can audit exactly what the app does with your microphone input, verify that no network requests are made during transcription, and confirm the architecture matches what we're describing here.
If you're in a regulated environment legal, medical, financial and your compliance team needs to review the code before approving the tool, they can. That's not possible with closed-source cloud tools.
What about the model download?
Tellaflow does make one network request: the initial download of the AI model (Whisper or Parakeet) when you first set it up. That download goes to a static file server the model weights are just files, not audio.
After that download, Tellaflow never makes another network request during normal use. You can disconnect from the internet and everything continues working exactly the same.
The practical implications
This architecture matters for specific situations:
- Attorneys: Attorney-client privilege extends to conversations and documents. Dictating client notes into a cloud tool introduces a third party into that privileged communication. On-device processing avoids this entirely.
- Healthcare: HIPAA-regulated environments have strict rules about PHI (protected health information). Audio of patient-related dictation in a cloud system is PHI. Local processing keeps it off covered entity servers.
- Security engineers: Air-gapped environments exist because some information should never touch external networks. A voice tool that requires internet breaks this guarantee. Tellaflow doesn't.
- Founders and executives: Your conversations about unreleased products, personnel decisions, and business strategy are worth protecting beyond what a vendor's privacy policy provides.
The principle behind it
We think privacy should be a property of the architecture, not a policy. Policies can change. Vendors get acquired. Servers get breached. The only reliable privacy guarantee is a system that physically cannot expose your data because it never has it.
That's what we built. Not "we promise to protect your audio." Just: your audio never leaves your Mac. That's the whole thing.