What Resonant Doesn't Do With Your Voice

Most voice dictation apps send your audio to a remote server. Your voice — your medical notes, legal briefs, private messages, proprietary code — travels through infrastructure you don't control.

Some apps say they don't store it. Some do. Some gate privacy behind enterprise pricing. We took a different approach: we made it structurally impossible for us to access your data in the first place.

What Resonant doesn't collect

Resonant processes speech recognition entirely on your Mac using on-device AI models. There is no server, no cloud endpoint, no WebSocket stream. Your audio never leaves your machine.

This means we don't collect:

• Your voice recordings. Audio is processed in real-time by local models and immediately discarded after transcription. We never receive, store, or have access to your voice.
• Your transcripts. The text output from your dictation stays on your device. We have no mechanism to retrieve it.
• Your screen context. Resonant does not read your screen, your clipboard, your active window text, or any application context. Some competitors use “context awareness” to scrape the text on your screen — we don't.
• Usage telemetry on your content. We don't track what you dictate, which words you use, or what apps you dictate into.

Does Resonant use my voice for AI training?

No. This is not a policy decision — it's an architectural one.

We cannot train on your voice data because we never have access to it. There is no “Privacy Mode” toggle because privacy is not optional. There is no enterprise tier required to unlock data protection. Every user gets the same guarantee: your data stays on your device.

For comparison, several cloud-based voice dictation services reserve “Zero Data Retention” for enterprise plans. Standard users on those platforms contribute their voice and transcripts to model training under broad “Service Improvement” clauses. We think that's the wrong tradeoff.

The Proton model: privacy through architecture

We take direct inspiration from Proton — the organization behind ProtonMail, ProtonVPN, and Proton Drive. Proton's philosophy is that privacy should be enforced by engineering, not by policy.

ProtonMail can't read your emails because they're end-to-end encrypted before Proton's servers ever see them. The privacy guarantee is structural. Even if Proton were compelled by a court order, they cannot produce data they don't have.

Resonant applies the same principle to voice. We can't listen to your dictation because the audio never leaves your Mac. We can't hand over transcripts to a third party because we don't have them. We can't train a model on your voice because the data never reaches our infrastructure.

This isn't a premium feature. It's the foundation.

How it works technically

Resonant bundles state-of-the-art speech recognition models that run directly on Apple Silicon. The Neural Engine in M-series chips provides the compute needed to run these models at speeds that rival cloud services — without a network connection.

When you speak:

1. Audio is captured by your Mac's microphone.
2. The on-device model processes the audio in real-time.
3. Text appears in your active application.
4. The audio buffer is immediately discarded.

No step in this process involves a network request. Resonant works offline, on airplanes, and in air-gapped environments. The only network activity is optional: license validation and crash reporting (which never includes voice data or transcription content).

Who this matters for

If you work with sensitive information, the architecture of your tools matters as much as their features.

• Attorneys drafting privileged communications need assurance that their dictation isn't traversing third-party infrastructure.
• Healthcare professionals documenting patient encounters need tools that don't create HIPAA liability through cloud transmission.
• Journalists protecting sources cannot afford voice data sitting on a vendor's server, subject to subpoena.
• Software engineers dictating proprietary logic and code need certainty that their IP won't end up in a training corpus.
• Therapists and counselors taking session notes handle some of the most sensitive content imaginable.

For all of these users, Resonant offers something most cloud services cannot: the absence of data — not the promise of its protection, but the structural impossibility of its collection.

The privacy hierarchy in voice dictation

Not all privacy claims are equal. Based on architecture, voice dictation tools fall into a clear hierarchy:

We believe the first tier is the only one that provides a genuine guarantee. Contracts can be amended. Policies can change. Servers can be breached. But data that never leaves your device cannot be compromised at the vendor level.

Frequently asked questions