Private dictation for macOS

Speak.
Stay in flow.

Turn your voice into clean text directly on your Mac. Fast, private, and without cloud inference.

Apple SiliconLocal inferenceNo API key

SVARA / PRODUCT FILM

LIVE LOCALLY
⌘ ⌥ DSpeak and keep moving
Scroll to explore

The shortcut

One shortcut. One thought. Done.

Svara gets out of the way. Focus a field, speak, and return to the work already in front of you.

SVARA / LOCAL WORKER•••
Ready for a shortcut
The best ideas arrive before the cursor is ready.Speak to fill this space
Dto dictate

From voice to text

Your ideas should move as fast as you do.

The waveform compresses. The thought stays.

Local by design

Your voice never needs to leave your Mac.

After setup, Svara runs speech recognition locally. No cloud inference. No API keys. No transcript history.

LOCAL ON THIS MAC
Microphoneopens only while dictating
Local workerstays ready on your Mac
Faster Whisper Tinyruns without a network
Clean textpastes into your active field
Offline after model installation Clipboard fallback included

A quieter way to write

Small details. A much better flow.

Svara is opinionated about the moments that matter: less setup, less context switching, more of your actual thought.

01

Local by design

Speech recognition runs on your Mac after setup.

MicWorkerText
02

Shortcut first

Press ⌘ ⌥ D and start speaking.

D
03

Personal dictionary

Keep names, tools, and terms formatted your way.

swaraSvaramac osmacOS
04

Automatic paste

Your result returns to the field you were already using.

Ideas move with you
05

Mic only when needed

The microphone opens only during active dictation.

active only while speaking
06

No account required

No API key. No cloud account. No transcript history.

Performance

Built to feel immediate.

Measured on Apple M4 with 16 GB RAM. The model stays ready, so your next thought does not wait for a reload.

Local measurement / Apple M4
0.085–0.104sShortcut to microphone active
0.22–0.55sTiny inference after recording
0.34–0.67sStop to final Clean result
Whisper load per worker session

Local measurements. Performance varies by hardware, audio length, and system state.

The persistent worker

The model stays ready. The microphone doesn’t.

Svara keeps Faster Whisper Tiny resident in a persistent local worker, so each dictation does not reload the model. The microphone still opens only when you explicitly start dictating.

01APP LAUNCHOne quiet setup
02LOAD TINY ONCEReady in the background
03READYWaiting for ⌘ ⌥ D
PERSISTENT WORKERFaster Whisper Tiny stays resident
⌘⌥DRecordTranscribePaste
⌘⌥DRecordTranscribePaste
⌘⌥DRecordTranscribePaste

Personal dictionary

Your words.
Your way.

Define the terms you use most. Svara applies them locally during cleanup, so names and tools come out right the first time.

PERSONAL DICTIONARY LOCAL
swaraSvara
pytorchPyTorch
mac osmacOS

In the flow

See Svara in flow.

One shortcut, from the first word to the final paste.

Svara / product filmLocal by default

Open source

Built in the open.

Svara is a small, focused stack with the local boundary in plain sight.

Desktop
01Tauri
02Rust
03React
04TypeScript
Speech / ML
01Faster Whisper
02CTranslate2
03WebRTC VAD
Backend
01Python
View source on GitHub

Stop typing every thought.
Start speaking.

Download Svara for Apple Silicon Macs.

v0.1.0Apple SiliconTiny model downloaded during setup

Current builds are not yet Apple notarized.