Skip to content
WebLLM

Data & Privacy

Privacy-First Design

Section titled “Privacy-First Design”

WebLLM is designed with privacy as a core principle. Users maintain full control over their data, provider choices, and retention policies.

Data Storage

Section titled “Data Storage”

What Gets Stored

Section titled “What Gets Stored”

The extension stores locally (in IndexedDB):

  • Conversation history - Requests and responses
  • Provider configurations - API keys, priorities
  • Permissions - Which sites can use WebLLM
  • Settings - User preferences

What Doesn’t Get Stored

Section titled “What Doesn’t Get Stored”
  • ❌ No data sent to extension developers
  • ❌ No analytics or tracking
  • ❌ No third-party services
  • ❌ No cloud sync (unless user explicitly enables)

Storage Location

Section titled “Storage Location”

All data stored in:

  • IndexedDB - Browser’s local database
  • Encrypted at rest - API keys encrypted
  • Per-profile - Separate for each browser profile
  • User-controlled - Can export or delete anytime

Data Retention Policies

Section titled “Data Retention Policies”

Users configure how long to keep history:

Retention Options

Section titled “Retention Options”
Data Retention Settings:
├── Keep history: [7 days | 30 days | 90 days | Forever]
├── Auto-delete after expiration: [✓]
└── Clear all data now [Button]

Automatic Cleanup

Section titled “Automatic Cleanup”

When auto-delete enabled:

  • Runs daily at midnight
  • Deletes conversations older than retention period
  • Keeps settings and permissions
  • Notification shown after cleanup

Manual Cleanup

Section titled “Manual Cleanup”

Users can:

  • Clear all conversation history
  • Delete specific conversations
  • Remove provider configurations
  • Revoke all permissions
  • Reset extension to defaults

Privacy Modes

Section titled “Privacy Modes”

1. Local-Only Mode

Section titled “1. Local-Only Mode”

Highest privacy - Use only local models:

  • ✅ Data never leaves browser
  • ✅ No internet required (after model download)
  • ✅ Zero cost
  • ✅ No API keys needed

Limitations:

  • Smaller models (1-4B parameters)
  • Slower inference
  • Limited capabilities

2. Personal API Keys

Section titled “2. Personal API Keys”

High privacy - Use your own API accounts:

  • ✅ Direct connection to provider
  • ✅ No intermediaries
  • ✅ You control API key
  • ✅ You see API usage

Data considerations:

  • Prompt data sent to AI provider
  • Subject to provider’s privacy policy
  • You pay directly

3. Application-Provided API

Section titled “3. Application-Provided API”

Standard privacy - App provides API access:

  • ⚠️ Application sees requests
  • ⚠️ Subject to app’s privacy policy
  • ✅ Convenient (no setup)
  • ✅ Often free for users

Data Flow Transparency

Section titled “Data Flow Transparency”

Usage Notifications

Section titled “Usage Notifications”

Every time a site uses WebLLM:

[WebLLM Icon] example.com is using AI
Action: Summarizing article
Provider: Local Model (Llama 3.2)
[View Details] [Block Site]

Notification shows:

  • Which site is making request
  • What action (summarize, translate, etc.)
  • Which provider is being used
  • Option to block immediately

Permission Prompts

Section titled “Permission Prompts”

First time a site uses WebLLM:

Allow example.com to use WebLLM?


This site wants to:
• Summarize articles
• Extract information
• Translate text


[Always Allow] [Allow Once] [Block]

Audit Log

Section titled “Audit Log”

View complete history in extension:

Recent Activity:
├── example.com - Summarize - 2m ago - Local Model
├── docs.site.com - Translate - 5m ago - Anthropic
└── app.site.com - Extract - 10m ago - OpenAI


[Export Log] [Clear History]

Security Measures

Section titled “Security Measures”

API Key Protection

Section titled “API Key Protection”
  • Encrypted storage - Keys encrypted in IndexedDB
  • Never exposed - Not accessible to web pages
  • Secure input - Masked in UI
  • Revocable - Delete anytime

Origin Isolation

Section titled “Origin Isolation”
  • Per-origin permissions - Each site separately approved
  • No cross-origin access - Sites can’t see each other’s requests
  • Sandboxed execution - Content script runs isolated

Request Validation

Section titled “Request Validation”

Before processing any request:

  1. ✓ Validate origin has permission
  2. ✓ Check request size limits
  3. ✓ Sanitize inputs
  4. ✓ Rate limit per origin
  5. ✓ Timeout enforcement

No Tracking

Section titled “No Tracking”

The extension:

  • ❌ Doesn’t collect analytics
  • ❌ Doesn’t phone home
  • ❌ Doesn’t share data with developers
  • ❌ Doesn’t use third-party services
  • ✅ 100% local operation

GDPR Compliance

Section titled “GDPR Compliance”

WebLLM extension is designed to help with GDPR compliance:

Data Minimization

Section titled “Data Minimization”
  • Only stores what’s necessary
  • Automatic deletion policies
  • User-controlled retention

Right to Access

Section titled “Right to Access”
  • Export all stored data as JSON
  • View complete audit log
  • Inspect provider configurations

Right to Erasure

Section titled “Right to Erasure”
  • One-click data deletion
  • Removes all conversation history
  • Clears all permissions

Right to Portability

Section titled “Right to Portability”
  • Export conversation history
  • Export settings
  • Import to another profile/browser

Privacy Comparison

Section titled “Privacy Comparison”

Traditional AI Integration

Section titled “Traditional AI Integration”
User → Website → Website Server → AI Provider
       ↑────────────────────────↑
       Website sees everything
  • ❌ Website sees all prompts
  • ❌ Website pays (or charges you)
  • ❌ No control over provider
  • ❌ Data stored on website’s servers

WebLLM

Section titled “WebLLM”
User → Website → Browser Extension → AI Provider
                 ↑──────────────────↑
                 User controls everything
  • ✅ Website only sees results (if you allow)
  • ✅ You control API keys/costs
  • ✅ You choose provider
  • ✅ Data stored locally or sent to provider directly

Best Practices

Section titled “Best Practices”

For Users

Section titled “For Users”
  1. Use local models when possible - Maximum privacy
  2. Review permissions regularly - Revoke unused access
  3. Set retention policies - Don’t keep data forever
  4. Monitor activity - Check audit log periodically
  5. Use personal API keys - More privacy than app-provided

For Developers

Section titled “For Developers”
  1. Minimize data collection - Only request what’s needed
  2. Be transparent - Explain why you need AI
  3. Respect user choices - Handle denial gracefully
  4. Don’t log prompts - If user uses their API key, respect it
  5. Provide fallbacks - Work without WebLLM if unavailable

Privacy FAQ

Section titled “Privacy FAQ”

Q: Can websites see my API keys? A: No. API keys never leave the extension and are not accessible to web pages.

Q: Can the extension developers see my data? A: No. Everything is stored locally. No data is sent to us.

Q: What does the AI provider see? A: When using API providers, they see the prompts sent to them (subject to their privacy policy). When using local models, nothing leaves your browser.

Q: Can I use WebLLM offline? A: Yes, if you use local models. Download the model once, then it works offline.

Q: How do I delete all my data? A: Extension Settings → Data Retention → “Clear all data now”

Q: Can websites track my usage? A: Websites only see what you explicitly allow. The extension doesn’t share usage patterns between sites.

Next Steps

Section titled “Next Steps”