Vector search solution overview

engine

The vector search feature in the engine is built on a highly optimized, proprietary version of pgvector and is fully compatible with standard pgvector usage. Key capabilities:

• Hybrid search across multimodal data, including structured data, graphs, and full-text search.

• Linear performance scaling by adding read-only nodes (RO nodes), supporting tens of thousands of QPS.

• Multiple quantization algorithms — PQ, SQ4, SQ8, and RabitQ — for up to 32x memory compression with high recall.

• Vector search on datasets from millions to billions of vectors.

For details, see quantization algorithms and scalar filtering.

PolarSearch distributed vector search engine

PolarSearch is a dedicated search engine compatible with the Elasticsearch and OpenSearch ecosystems. Add a PolarSearch node to your cluster to run hybrid vector and full-text search at scale. Key capabilities:

• Specialized hybrid search combining full-text and vector search.

• Distributed high-performance search on datasets up to tens of billions of high-dimensional vectors.

• Quantization algorithms with multiple compression levels plus on-demand vector loading to balance performance and memory cost.

• Flexible ingest pipelines and search pipelines for advanced ML computation, including automatic embedding and reranking.