> ## Documentation Index
> Fetch the complete documentation index at: https://docs.voicetypr.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Model Management

> Download, switch, and manage AI models for offline transcription

VoiceTypr uses local AI models for offline transcription. Learn how to download, manage, and optimize your models.

## Understanding Models

VoiceTypr supports multiple AI model engines:

<CardGroup cols={3}>
  <Card title="Whisper" icon="w">
    OpenAI's speech recognition model

    **99+ languages**

    5 sizes available
  </Card>

  <Card title="Parakeet" icon="p">
    NVIDIA's optimized model (macOS only)

    **25 languages**

    Apple Neural Engine
  </Card>

  <Card title="Cloud Models" icon="cloud">
    Soniox and other cloud providers

    **API required**

    No download needed
  </Card>
</CardGroup>

## Model Characteristics

Each model shows three key metrics:

<Steps>
  <Step title="Speed Score">
    How fast the model transcribes (1-10)

    Higher = faster transcription

    <Icon icon="zap" /> Speed indicator
  </Step>

  <Step title="Accuracy Score">
    How accurate the transcription is (1-10)

    Higher = better accuracy

    <Icon icon="check-circle" /> Accuracy indicator
  </Step>

  <Step title="Size">
    Disk space required for the model

    Ranges from \~75 MB to \~2.9 GB

    <Icon icon="hard-drive" /> Size indicator
  </Step>
</Steps>

### Model Sizes Explained

| Model      | Size     | Speed | Accuracy | Best For                              |
| ---------- | -------- | ----- | -------- | ------------------------------------- |
| **Tiny**   | \~75 MB  | 10    | 4        | Quick testing, limited accuracy needs |
| **Base**   | \~142 MB | 9     | 5        | Fast transcription, casual use        |
| **Small**  | \~466 MB | 7     | 7        | Balanced performance                  |
| **Medium** | \~1.5 GB | 5     | 8        | High accuracy needed                  |
| **Large**  | \~2.9 GB | 3     | 10       | Maximum accuracy, professional use    |

<Info>
  <Icon icon="star" /> **Recommended models** are marked with a star icon and offer the best balance of speed and accuracy.
</Info>

## Downloading Models

<Steps>
  <Step title="Open Models Tab">
    Navigate to the Models section in VoiceTypr
  </Step>

  <Step title="Choose a Model">
    Browse "Available to Set Up" section

    Review speed, accuracy, and size metrics
  </Step>

  <Step title="Download">
    Click the "Download" button

    Progress is shown with a percentage bar
  </Step>

  <Step title="Wait for Verification">
    After download, model is verified (SHA-256 checksum)

    "Verifying..." indicator appears briefly
  </Step>

  <Step title="Model Ready">
    Model moves to "Available to Use" section

    Now selectable for transcription
  </Step>
</Steps>

<Tip>
  Start with a **Small** model for balanced performance. You can download multiple models and switch between them.
</Tip>

### Download Progress

While downloading:

* **Progress bar** shows download percentage
* **Cancel button** (X) allows canceling the download
* **Verifying state** appears after download completes

<Note>
  **Parakeet models** may show "Downloading..." without a percentage. This is normal - FluidAudio (the download system) doesn't report progress for these models.
</Note>

## Switching Models

Change your active transcription model at any time:

1. Go to Models tab
2. Find your desired model in "Available to Use"
3. Click anywhere on the model card
4. Active model is highlighted with a colored background

<Warning>
  Only one model can be active at a time. Switching models takes effect immediately for new recordings.
</Warning>

## Deleting Models

Free up disk space by removing unused models:

<Steps>
  <Step title="Locate Model">
    Find the model in "Available to Use" section
  </Step>

  <Step title="Click Remove">
    Click the "Remove" button on the model card
  </Step>

  <Step title="Confirm Deletion">
    Confirm you want to delete the model
  </Step>

  <Step title="Model Removed">
    Model moves back to "Available to Set Up"

    Disk space is freed immediately
  </Step>
</Steps>

<Warning>
  If you delete your currently active model, you'll need to select a different model before recording again.
</Warning>

## Model Verification

VoiceTypr ensures model integrity:

* **SHA-256 checksum** verification after download
* Corrupted downloads are detected automatically
* Failed verification prompts re-download

<Info>
  Verification typically takes 5-10 seconds depending on model size.
</Info>

## Storage Management

### Where Models are Stored

Models are stored in platform-specific locations:

* **macOS:** `~/Library/Application Support/com.voicetypr.app/models/`
* **Windows:** `%APPDATA%/com.voicetypr.app/models/`

### Checking Disk Usage

Total space used by all models:

* Tiny only: \~75 MB
* Tiny + Small: \~541 MB
* Small + Medium: \~1.97 GB
* All models: \~5 GB+

<Tip>
  Keep only the models you use regularly. You can always re-download models later.
</Tip>

### Cleaning Up

To free disk space:

1. Delete unused models (see above)
2. Keep 1-2 models that fit your needs
3. Advanced users: Manually delete from file system (not recommended)

## Cloud Models

Cloud-based models (like Soniox) don't require downloads:

<Steps>
  <Step title="Add API Key">
    Click "Add API Key" on the cloud model card
  </Step>

  <Step title="Enter Credentials">
    Provide your API key from the provider

    Keys are stored securely in system keychain
  </Step>

  <Step title="Model Ready">
    Model moves to "Available to Use"

    Select to use for transcription
  </Step>
</Steps>

<Note>
  Cloud models require an internet connection for transcription. See provider documentation for pricing and limits.
</Note>

### Removing Cloud Models

Click "Remove API Key" to disconnect:

* API key is deleted from keychain
* Model moves back to "Available to Set Up"
* No local data is deleted

## Best Practices

<AccordionGroup>
  <Accordion title="Choose the Right Model" icon="brain">
    * **Casual use:** Small model (best balance)
    * **Professional transcription:** Large model
    * **Quick notes:** Tiny or Base model
    * **Limited disk space:** Tiny or Base model
  </Accordion>

  <Accordion title="Download Over Wi-Fi" icon="wifi">
    Models range from 75 MB to 2.9 GB. Use a stable internet connection for downloading.
  </Accordion>

  <Accordion title="Preload Before Offline Use" icon="download">
    Download models while you have internet access. They work completely offline once downloaded.
  </Accordion>

  <Accordion title="Test Multiple Models" icon="flask">
    Try different models to find the best speed/accuracy balance for your use case.
  </Accordion>
</AccordionGroup>

## Troubleshooting

<AccordionGroup>
  <Accordion title="Download Fails" icon="exclamation-triangle">
    **Solution:**

    * Check internet connection
    * Ensure sufficient disk space (3-4 GB free)
    * Try canceling and restarting the download
  </Accordion>

  <Accordion title="Verification Fails" icon="shield">
    **Solution:**

    * Delete the model and re-download
    * Check disk for errors
    * Contact support if issue persists
  </Accordion>

  <Accordion title="Model Won't Select" icon="mouse-pointer">
    **Solution:**

    * Ensure model is fully downloaded and verified
    * Check that model appears in "Available to Use"
    * Restart VoiceTypr if needed
  </Accordion>
</AccordionGroup>

<Tip>
  If you encounter persistent issues, try the "Reset App Data" option in Advanced settings (this will delete all models and settings).
</Tip>