Local Offline Mode

Overview

Typeoff's local mode uses a speech recognition model running entirely on your device. No audio data is ever sent to the internet — everything stays on your machine.

Cloud vs. Local

Setting Up Local Mode

1. Download the Model

Before using local mode, you need to download the speech recognition model (~894 MB).

1. Open Typeoff Settings → Recognition
2. Under "Recognition Engine", select Local
3. Click Download Model
4. Wait for the download to complete — progress is shown in real-time

The model is stored in your app data directory and only needs to be downloaded once.

2. Switch to Local Mode

Once the model is downloaded, select Local as your recognition engine in Settings → Recognition. Typeoff will use local transcription for all future recordings.

3. Verify the Model

If transcription quality seems off, you can verify the model integrity:

1. Go to Settings → Recognition
2. Click Verify Model
3. If the model is corrupted, you'll be prompted to re-download

Supported Languages

Local mode supports the following languages:

• Chinese (Simplified)
• English
• Japanese
• Korean
• Cantonese

If you had other languages selected, they will be automatically filtered when switching to local mode. If none of your selected languages are supported, the language will default to automatic detection.

Performance

• Hardware acceleration: CoreML on macOS for faster inference
• Model: Optimized for on-device inference (ONNX Runtime)
• Latency: Near-instant after recording ends

When to Use Local Mode

• You work in environments without internet access
• You handle sensitive or confidential information
• You want zero-latency transcription
• You prefer not to send audio data to any server

Automatic Fallback

When using cloud mode, Typeoff automatically falls back to local mode if the internet connection is unavailable — provided the local model is downloaded.