lancedb

An embedded vector database integrates directly into your application for RAG pipelines without server deployments. LanceDB provides zero-config vector storage, allowing you to store embeddings locally and retrieve relevant information efficiently using semantic search.

Defining schemas for vector storage involves using Pydantic or Arrow to structure your data tables. This approach allows you to specify text fields for content and vector fields for embeddings, ensuring structured data management within your embedded database.

Yes, you can combine full-text search and semantic search using hybrid search with rerankers. This technique merges traditional keyword matching with vector-based retrieval, allowing the system to rerank results and improve search accuracy for complex queries.

Yes, an embedded serverless vector database supports data versioning to track changes in your stored embeddings over time. This capability ensures you can manage updates to your datasets and maintain historical states for your retrieval pipelines.

You filter metadata alongside vector search results by applying scalar indexing to your structured fields. This process narrows down the vector search space based on specific metadata attributes before performing semantic retrieval, optimizing query efficiency.