Vector Search
Anava Vector Search enables AI-powered semantic search across your video analytics data. Instead of rigid keyword matching, you can search using natural language queries like "busy cafe scene" or "person walking near entrance."
Key Capabilities
| Feature | Description |
|---|---|
| Semantic Understanding | AI understands context and meaning, not just text matching |
| Multimodal Search | Search by text description or find similar images |
| Hybrid Results | Combines text and image similarity using RRF fusion |
| Cost Effective | ~$4-5/month using serverless BigQuery |
How It Works
Quick Start
- Navigate to Vector Search in the sidebar
- Enter a natural language query in the search box
- Results appear ranked by relevance with thumbnails
Example Queries
| Query | What It Finds |
|---|---|
person walking near entrance | Semantic match on walking activity near entry points |
busy lunch hour | High-traffic scenes with "lunch" context |
delivery truck | Vehicles matching delivery truck appearance |
empty parking lot | Low-activity parking area scenes |
Architecture Overview
Vector Search uses a BigQuery-native pipeline that ingests events, generates embeddings, and performs similarity search with low-latency results. See the architecture page for the system overview.
Cost Breakdown
| Component | Monthly Cost |
|---|---|
| Multimodal Embeddings | ~$1.50 |
| BigQuery Queries | ~$2.00 |
| BigQuery Storage | ~$0.50 |
| BigQuery Streaming | ~$0.50 |
| Total | ~$4-5/month |
Based on 10,000 events/day, 100 queries/day
Security
- Firebase Authentication required for all queries
- Anti-discrimination pattern blocking
- Rate limiting per user
- All data stored in your GCP project
Next Steps
- Architecture Details - System overview and design goals
- Security Model - Enterprise security overview