- Vector Search Engine - Primary Algorithm
- HNSW (Hierarchical Navigable Small World) in k-NN plugin
- Vector Search Engine - Supported Vector Dimensions
- 384 (BERT base), 768, 1024, up to 4096+ custom
- Vector Search Engine - Distance Metrics
- Cosine similarity, Euclidean (l2), inner product
- Vector Search Engine - Maximum Vector Capacity
- Billions of vectors distributed across clusters
- Keyword Search Technology - Ranking Algorithm
- BM25 for hybrid search integration
- Keyword Search Technology - Tokenization
- Language-aware including multilingual analyzers
- Keyword Search Technology - Query Syntax
- DSL with neural, knn, bool queries; wildcards, phrases
- Embedding Model Support - Pre-trained Models
- Hugging Face BERT, Sentence-BERT via ML Commons
- Embedding Model Support - Custom Model Support
- TorchScript, ONNX, custom trained embedding models
- Embedding Model Support - Model Inference
- Local node inference, remote connectors, GPU support
- Infrastructure Requirements - Deployment Options
- AWS managed, self-hosted OSS, Kubernetes, Docker
- Infrastructure Requirements - Memory Per 1M Vectors
- ~400MB for 384-dim HNSW (configurable ef_construction)
- Infrastructure Requirements - High Availability
- Multi-node clusters, shard replication, automated failover
- Infrastructure Requirements - GPU Support
- GPU acceleration via plugins for model inference
- Scalability Limits - Maximum Document Count
- Trillions across distributed clusters
- Scalability Limits - Concurrent Queries
- 10,000+ QPS horizontally scalable
- Scalability Limits - Index Update Frequency
- Real-time per document via semantic fields
