Skip to content
Secretary42
Offline dictation

Offline dictation without cloud transcription

Secretary42 records and transcribes your speech on your computer. Once a Whisper model is installed and the licence is active, ordinary dictation does not need an internet connection and cannot silently reroute your audio to a cloud provider.

Reviewed by the Secretary42 product team · Updated 27 August 2026

What “offline dictation” means here

Press the global hotkey, speak, and stop recording. Secretary42 converts the captured audio to the format required by whisper.cpp, runs the selected model locally, and sends the result to the active field, clipboard, or file. Microphone audio and transcript content remain on the device throughout this path.

Secretary42 recording state on macOS
Recording is explicit and can be stopped immediately.
Secretary42 local transcription state on macOS
The selected Whisper model transcribes after recording stops.

What still uses the internet

“Offline transcription” is deliberately narrower than “the application never connects.” A one-time activation and silent daily best-effort lease refresh contact the EU licensing service using pseudonymous licensing data. Model downloads, update checks, Paddle checkout, support, and diagnostics you opt into are also online operations. None of those operations carries your microphone audio or transcript text.

Why a strict failure path matters

Some products use local speech recognition only as a preference and switch to a remote provider if it fails. Secretary42’s direct desktop build does not. If the model, local engine, or audio conversion fails, the app reports that local failure and stops. This makes the privacy boundary easier to understand and verify.

Before working offline

  1. Activate your licence while connected to the internet.
  2. Download and prepare the Whisper model you want to use.
  3. Run a short dictation test in the target application.
  4. Confirm the selected output mode: active field, clipboard, or file.

What Secretary42 does not do

  • No cloud-transcription fallback and no provider API key is needed.
  • No account sync, shared transcript library, or browser dashboard.
  • No AI rewriting, agent commands, or multi-speaker meeting transcription.

Use speech-to-text with a clear privacy boundary

Secretary42 is available for macOS on Apple Silicon and Windows x64.

View downloads