Resonant — Private dictation for Mac
Privacy is architecture, not policy. Resonant runs speech recognition entirely on your hardware. There is no server to send your audio to. No account to create. No internet connection required.
On-device AI. Apple Neural Engine. Audio discarded after transcription.
What private actually means
Most dictation apps promise privacy in their terms of service. Resonant enforces it in the code. There is no upload path, no telemetry endpoint, no analytics pipeline for your audio.
Your voice is processed entirely on your Mac. Audio never touches a server, a CDN, or an API endpoint. There is no server to send it to.
Download, open, dictate. No email, no sign-up form, no OAuth flow. The app works the moment you launch it.
Resonant runs on-device speech recognition. Disconnect your Wi-Fi, turn off Bluetooth, pull the ethernet cable. It still works.
Audio is discarded the instant transcription completes. The raw waveform exists in memory for the duration of processing and nowhere else.
Your dictations never contribute to model improvement, fine-tuning, or analytics. The models run locally and learn nothing from your usage.
On-device architecture
Resonant uses NVIDIA Parakeet v3 and Qwen3 ASR models compiled for Apple Neural Engine via CoreML. Speech recognition runs entirely on-device — no cloud fallback, no hybrid mode, no “processing locally when possible.”
Audio exists in memory only during transcription. The moment text is produced, the raw waveform is released. No temp files, no cache, no buffer written to disk.
Microphone audio streams into a memory buffer. Nothing touches the filesystem.
NVIDIA Parakeet v3 or Qwen3 ASR runs on Apple Neural Engine. Sub-second latency, fully local.
On-device language model corrects formatting, punctuation, and filler words. Still local.
Raw audio buffer is released from memory. Only the finished text remains.
Comparison
| Feature | Resonant | Cloud services |
|---|---|---|
| Audio leaves device | Never | Always |
| Account required | No | Yes |
| Works offline | Yes | No |
| Data retention | None — audio discarded after transcription | 30 days to indefinite |
| HIPAA mode | Built-in — encrypted storage, audit log | Enterprise tier or unavailable |
| Price | Free tier included | $10–30/mo |
Who needs this
Client conversations, case notes, and privileged communications stay on your machine. No third-party processor means no subpoena risk on someone else’s servers.
Patient notes, clinical observations, and treatment plans dictated without cloud exposure. HIPAA mode adds encrypted storage and audit logging.
Board prep, M&A discussions, and strategic planning captured by voice without creating a cloud trail. Your thoughts stay yours until you choose to share them.
Source conversations, investigative notes, and story drafts dictated without exposure to third-party infrastructure. Protect your sources by architecture, not policy.
Financial records, personal journals, proprietary research, HR documentation. If the content matters enough to think twice before typing it into a web app, dictate it locally instead.
HIPAA mode
HIPAA mode adds a layer of operational controls on top of Resonant's already-local architecture. Everything that was private stays private — HIPAA mode makes it auditable.
All transcriptions encrypted at rest using macOS Keychain-managed keys. Data is unreadable outside of Resonant.
Every transcription event is logged — timestamp, duration, word count. Exportable for compliance review. No content is logged.
All optional analytics and crash reporting disabled. Zero outbound data. The only network traffic is update checks, which can also be disabled.
Don't take our word for it
Open Activity Monitor or any network inspector. Launch Resonant. Dictate anything — your most sensitive thoughts, your most confidential notes. Watch the network tab.
Zero outbound connections.
Try it
nettop -p Resonant -dThis isn't a privacy policy you have to read. It's an architecture you can inspect. If Resonant ever phones home, your network monitor will catch it. It won't.
Free. Local. Verifiable.
On-device speech recognition. No cloud, no account, no data retention. Just your voice and your Mac.
Requires macOS 14+ · Apple Silicon