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

# Privacy-First Setup

> Configure Stenox for 100% local, offline operation with complete privacy

"og:title": "Stenox Docs"

## Complete Privacy with Local Processing

Stenox can operate completely offline using local models for both transcription and AI enhancement. This means your audio and text never leave your Mac - perfect for sensitive content, confidential work, or privacy-conscious users.

<Tip>
  **100% Private = WhisperKit (transcription) + MLX (AI enhancement)**

  No internet connection required, no API keys, no cloud services, complete privacy.
</Tip>

## Why Privacy-First?

<CardGroup cols={2}>
  <Card title="Complete confidentiality" icon="lock">
    Healthcare, legal, financial, or any sensitive content.
  </Card>

  <Card title="Offline functionality" icon="wifi-slash">
    Works on airplanes, remote locations, or with no internet.
  </Card>

  <Card title="No third parties" icon="shield-check">
    Your data never sent to external services.
  </Card>

  <Card title="No API costs" icon="dollar-sign">
    No per-minute charges or external API fees.
  </Card>
</CardGroup>

## Privacy-First Setup Guide

Follow these steps to configure Stenox for completely private operation.

### Step 1: Configure WhisperKit (Local Transcription)

<Steps>
  <Step title="Open Stenox Settings">
    Click the Stenox menu bar icon → **Settings**
  </Step>

  <Step title="Go to Models tab">
    Select the **Models** tab
  </Step>

  <Step title="Select WhisperKit">
    Under **Transcription Provider**, choose **WhisperKit**
  </Step>

  <Step title="Choose a model">
    Select your preferred model:

    * **base.en** (142 MB) - Fastest, English only
    * **small.en** (483 MB) - Better accuracy, English only
    * **base** (142 MB) - Multilingual (100+ languages)
    * **small** (483 MB) - Better accuracy, multilingual

    **Recommended:** base.en for beginners
  </Step>

  <Step title="Download model">
    Click **Download** and wait for completion (\~142 MB or \~483 MB)

    Model saved to: `~/stenox-models/whisper/`
  </Step>

  <Step title="Verify offline mode">
    Turn off Wi-Fi and test a dictation - it should work perfectly!
  </Step>
</Steps>

[Learn more about WhisperKit →](/providers/transcription/whisperkit-local)

### Step 2: Configure MLX (Local AI Enhancement)

<Steps>
  <Step title="In the same Models tab">
    Under **AI Enhancement Provider**, select **MLX**
  </Step>

  <Step title="Choose a model">
    Select based on your Mac's specifications:

    * **Gemma 3n 2B** (\~1.2 GB) - Best accuracy, recommended
    * **Qwen 2.5 1.5B** (\~1 GB) - Fastest, 8GB RAM minimum
    * **Phi-3 Mini** (\~2.4 GB) - Balanced quality, 12GB RAM recommended
    * **Gemma 3n 4B** (\~2.5 GB) - Higher quality, 16GB RAM recommended

    **Recommended:** Gemma 3n 2B for best accuracy with fast processing
  </Step>

  <Step title="Download model">
    Click **Download** and wait for completion (\~1-2.5 GB)

    Model saved to: `~/stenox-models/mlx/`
  </Step>

  <Step title="Test enhancement">
    Perform a dictation - it will be transcribed AND enhanced locally
  </Step>
</Steps>

<Warning>
  **MLX requires Apple Silicon.** Intel Mac users cannot use MLX and should skip AI enhancement or use cloud providers.
</Warning>

[Learn more about MLX →](/providers/llm/mlx-local)

### Step 3: Verify Privacy Settings

<Steps>
  <Step title="Check active providers">
    In Settings, verify:

    * **Transcription:** WhisperKit
    * **AI Enhancement:** MLX (or None)
  </Step>

  <Step title="Remove cloud API keys (optional)">
    For complete privacy, remove any cloud API keys:

    * Go to **Transcription** tab → Remove DeepGram/AssemblyAI/Groq keys
    * Go to **AI Enhancement** tab → Remove Gemini/Groq keys
  </Step>

  <Step title="Test offline">
    1. Disconnect from internet (turn off Wi-Fi)
    2. Perform a dictation
    3. Verify transcription and enhancement work
  </Step>

  <Step title="Done!">
    Stenox now operates 100% locally with complete privacy.
  </Step>
</Steps>

## What You Get

### Complete Privacy

* **Audio stays local** - Voice recordings never uploaded
* **Text stays local** - Transcriptions processed on-device
* **No internet required** - Works offline
* **No logging** - Nothing tracked or stored externally
* **No third parties** - No cloud providers see your data

### No API Costs

* **No per-minute charges** - Local processing has no external fees
* **No API limits** - No third-party rate limits
* **No credit cards for providers** - No cloud provider accounts needed

### Full Functionality

* **All features work** - Transcription, enhancement, profiles, vocabulary
* **Customizable** - Create profiles for different use cases
* **Auto-activation** - Profiles switch based on active app

## Performance Expectations

### Processing Time

With local processing, expect these speeds:

| Configuration                             | Total Time     | Breakdown                             |
| ----------------------------------------- | -------------- | ------------------------------------- |
| **WhisperKit base.en only**               | \~3-5 seconds  | 3-5s transcription                    |
| **WhisperKit + MLX Gemma 3n 2B**          | \~5-7 seconds  | 3-4s transcription + 2-3s enhancement |
| **WhisperKit small.en + MLX Gemma 3n 4B** | \~8-10 seconds | 5-6s transcription + 3-4s enhancement |

<Note>
  Local processing is slower than cloud (which can be \~1-2 seconds total), but provides complete privacy.
</Note>

### Apple Silicon Note

<Note>
  **MLX requires Apple Silicon.** Intel Mac users can use WhisperKit for local transcription, but need cloud providers for AI enhancement.
</Note>

## Privacy Use Cases

<AccordionGroup>
  <Accordion title="Healthcare & Medical" icon="heart-pulse">
    **Why private:** HIPAA compliance, patient confidentiality

    **Setup:**

    * WhisperKit for medical dictation
    * MLX for note formatting
    * Custom vocabulary for medical terms
    * Profile for EMR/EHR applications
  </Accordion>

  <Accordion title="Legal Work" icon="scale-balanced">
    **Why private:** Attorney-client privilege, confidential documents

    **Setup:**

    * WhisperKit for case notes, briefs
    * MLX for professional formatting
    * Vocabulary for legal terminology
    * Profile for document management apps
  </Accordion>

  <Accordion title="Financial Services" icon="building-columns">
    **Why private:** Client confidentiality, regulatory compliance

    **Setup:**

    * WhisperKit for financial analysis
    * MLX for report formatting
    * Profile for Excel, financial software
  </Accordion>

  <Accordion title="Journalism" icon="newspaper">
    **Why private:** Source protection, sensitive investigations

    **Setup:**

    * WhisperKit for interview transcription
    * MLX for article drafting
    * Profiles for different publications
  </Accordion>

  <Accordion title="Personal Privacy" icon="user-lock">
    **Why private:** Personal preference, data sovereignty

    **Setup:**

    * WhisperKit for all dictation
    * MLX for text improvement
    * Complete offline operation
  </Accordion>
</AccordionGroup>

## Optional: Disable AI Enhancement

If you want the fastest local processing or don't need enhancement:

<Steps>
  <Step title="Open Settings → AI Enhancement tab">
    Go to the AI Enhancement settings
  </Step>

  <Step title="Select 'None'">
    Choose **None** as the AI enhancement provider
  </Step>

  <Step title="Save">
    Enhancement will be skipped, reducing processing time to \~3-5 seconds (WhisperKit only)
  </Step>
</Steps>

**Benefits:**

* Faster processing (3-5s vs 5-7s)
* Less RAM usage
* Still 100% private
* Raw transcription without formatting

## Advanced Privacy Configuration

### Custom Model Storage Location

Change where models are stored:

<Tabs>
  <Tab title="Default Location">
    Models stored in `~/stenox-models/`
  </Tab>

  <Tab title="Custom Location">
    1. Open Stenox Settings → **Models** tab
    2. Click **Change Model Directory**
    3. Select preferred location (e.g., external drive)
  </Tab>
</Tabs>

### Encrypted Disk Image (Advanced)

For maximum security, store models on encrypted disk image:

<Steps>
  <Step title="Create encrypted disk image">
    Use macOS Disk Utility to create an encrypted APFS image
  </Step>

  <Step title="Mount the image">
    Mount the encrypted disk and unlock with password
  </Step>

  <Step title="Change model directory">
    Point Stenox model directory to encrypted disk
  </Step>

  <Step title="Download models">
    Models are now stored encrypted at rest
  </Step>
</Steps>

## Troubleshooting

<AccordionGroup>
  <Accordion title="Models won't download" icon="cloud-arrow-down">
    * Models require internet to download (one-time)
    * After download, they work offline forever
    * Ensure sufficient disk space (2-5 GB)
  </Accordion>

  <Accordion title="Processing is too slow" icon="hourglass">
    * Use base.en instead of small.en (faster)
    * Use Qwen 2.5 1.5B or Gemma 3n 2B instead of larger models (faster)
    * Skip AI enhancement (None) for fastest processing
    * Consider hybrid approach (local transcription + cloud enhancement)
  </Accordion>

  <Accordion title="MLX not available (Intel Mac)" icon="ban">
    * MLX requires Apple Silicon
    * Options for Intel Macs:
      1. Use WhisperKit alone (no enhancement)
      2. Use cloud enhancement (Gemini/Groq)
      3. Upgrade to Apple Silicon Mac
  </Accordion>

  <Accordion title="Out of disk space" icon="hard-drive">
    * Delete unused models from `~/stenox-models/`
    * Use smaller models (base instead of small)
    * Skip AI enhancement to save 1-2.5 GB
  </Accordion>
</AccordionGroup>

## Privacy Checklist

✅ **Transcription:** WhisperKit (local)
✅ **AI Enhancement:** MLX (local) or None
✅ **Offline test:** Works with Wi-Fi off
✅ **Cloud API keys:** Removed or not entered
✅ **Internet not required:** All models downloaded
✅ **Verified:** No data sent to cloud

<Note>
  With this configuration, Stenox operates entirely on your Mac with zero external communication.
</Note>

## Hybrid Approach (Alternative)

If you want privacy for audio but cloud quality for text:

**Setup:**

* **Transcription:** WhisperKit (local) - Audio stays private
* **Enhancement:** Google Gemini or Groq (cloud) - Text enhanced in cloud

**Benefits:**

* Most sensitive data (audio) stays local
* High-quality enhancement via cloud
* Faster than full local (4-5s vs 7-10s)

[Learn about Hybrid Setup →](/guides/free-tier-providers#option-4-hybrid-privacy--quality)

## Cloud API Keys & Privacy Policies

When using cloud providers, Stenox supports bring-your-own-API-keys. This means **your data privacy is governed by your API provider's policies**, not Stenox.

<Info>
  **Your API, Your Rules:** Stenox never sees your API keys or data. All communication goes directly from your Mac to your chosen provider using your credentials.
</Info>

### Secure API Key Storage

<CardGroup cols={2}>
  <Card title="macOS Keychain Storage" icon="key">
    API keys are stored securely in your **macOS Keychain**, the same system that protects your passwords and certificates.
  </Card>

  <Card title="Single Purpose Only" icon="lock-keyhole">
    Keys are **only used to access your chosen provider**. Stenox has no other use for your API keys - no analytics, no tracking, no secondary purposes.
  </Card>

  <Card title="Never Transmitted" icon="ban">
    Your API keys never leave your Mac except to authenticate directly with your provider's API.
  </Card>

  <Card title="You Control Access" icon="user-shield">
    You can view, modify, or delete your API keys at any time through Stenox settings or macOS Keychain Access.
  </Card>
</CardGroup>

### Privacy Governance

Your API key's privacy settings are fully honored:

<CardGroup cols={2}>
  <Card title="Enterprise & Business Accounts" icon="building">
    If you have enterprise or business accounts with your cloud provider (DeepGram, AssemblyAI, Groq, Google, etc.), your organization's privacy policies are automatically applied.
  </Card>

  <Card title="Private Policy Attachments" icon="file-contract">
    Any private policies or data processing agreements attached to your API keys are fully honored and enforced by your provider.
  </Card>

  <Card title="Direct Communication" icon="arrow-right-arrow-left">
    Stenox sends data directly to your provider using your API key - no intermediary servers.
  </Card>

  <Card title="Provider Compliance" icon="shield-check">
    Your provider's security certifications (SOC 2, HIPAA, GDPR, etc.) apply to your API requests.
  </Card>
</CardGroup>

### How It Works

<Steps>
  <Step title="You provide API key">
    Enter your API key in Stenox settings
  </Step>

  <Step title="Stenox uses your credentials">
    All requests use your API key directly
  </Step>

  <Step title="Provider applies your policies">
    Your enterprise policies, privacy settings, and agreements are enforced
  </Step>

  <Step title="You maintain control">
    Data processing follows your provider account's rules
  </Step>
</Steps>

<Tip>
  **For maximum privacy with cloud quality:** Use business/enterprise API keys with your organization's data processing agreements in place.
</Tip>

## Next Steps

<CardGroup cols={2}>
  <Card title="WhisperKit Guide" icon="microphone" href="/providers/transcription/whisperkit-local">
    Complete guide to local transcription.
  </Card>

  <Card title="MLX Guide" icon="sparkles" href="/providers/llm/mlx-local">
    Complete guide to local AI enhancement.
  </Card>

  <Card title="Create Profiles" icon="sliders" href="/profiles/creating-profiles">
    Set up profiles for different privacy needs.
  </Card>

  <Card title="Vocabulary" icon="book" href="/setup/vocabulary">
    Add custom terms for better accuracy.
  </Card>
</CardGroup>

<Tip>
  **Perfect for sensitive work:** Privacy-first setup gives you complete control and confidentiality with zero cloud dependencies.
</Tip>