AI-Powered Search

Tamsaek includes a local AI model that understands natural language queries, time references, and contextual search.

How It Works

1. Local Model: A Qwen 2.5 language model runs entirely on your device
2. Query Understanding: AI parses natural language into structured search
3. Semantic Matching: Understands meaning, not just keywords
4. Hybrid Results: Combines AI with traditional full-text search

Privacy First

• No cloud: All processing happens locally
• No data sharing: Your files never leave your device
• Offline capable: Works without internet after initial model download
• One-time download: ~3GB model downloaded once

Natural Language Examples

Query	What AI Understands
”find the contract from last month”	File type: document, Time: past 30 days, Content: contract
”John’s presentation about sales”	Owner: John, Type: presentation, Topic: sales
”receipts from restaurants in 2024”	Type: receipt/invoice, Category: restaurant, Year: 2024

Enabling AI Search

1. Open Settings → AI
2. Toggle Enable AI Search
3. Wait for model download (one-time, ~3GB)
4. Press Tab in search bar to activate

AI Modes

Standard AI

Natural language understanding for queries:

find the budget spreadsheet I worked on last week

API Mode (Optional)

Use external AI providers for enhanced capabilities:

1. Settings → AI → API Mode
2. Enter API key (Groq, OpenAI, etc.)
3. Faster responses, more capable

System Requirements

For local AI:

• RAM: 8GB minimum, 16GB recommended
• Storage: 3GB for model
• CPU: Any modern processor (GPU not required)

Performance Tips

1. First query is slow: Model loads on first use (~10s)
2. Subsequent queries: Fast response times
3. Memory usage: Model uses 2-4GB RAM when active
4. Fallback: Use regular search for simple keyword queries