pgvector
pgvector is an open-source vector extension plugin for PostgreSQL that enables storing, indexing, and querying high-dimensional vector data within a relational database. It supports both exact search and approximate nearest neighbor (ANN) search, provides multiple distance metrics such as Euclidean distance, cosine similarity, and inner product, and is compatible with index algorithms including IVFFlat and HNSW. pgvector is commonly used for storing and retrieving machine learning embedding vectors, allowing developers to implement semantic search and recommendation features within PostgreSQL without a dedicated vector database.
Core Facts
The pgvector extension adds vector storage and similarity search capabilities to PostgreSQL, enabling Supabase to serve simultaneously as a traditional database and a vector database.
Pinecone、Chroma、Weaviate是常见的向量数据库产品
Long-term memory is typically implemented with the help of vector databases (such as Chroma, Pinecone, Weaviate, Milvus, etc.) to enable semantic retrieval.
Timeline (last 90 days)
For each query, chatVia.ai runs in parallel a pgvector-based dense vector similarity search and an in-Postgres sparse BM25 keyword search, and merges the rankings using Reciprocal Rank Fusion.
pgvector's performance in ultra-large-scale vector scenarios is inferior to that of dedicated vector databases.
OntoGr runs on top of FalkorDB or Neo4j, and can optionally use pgvector or Postgres; a background consolidation task handles deduplication, summarization, scoring, and archiving; the LLM and embedder are bring-your-own.
Dedicated vector databases (such as Pinecone, Qdrant, and Weaviate) offer more extreme performance optimization and distributed scaling for large-scale vector retrieval, while pgvector's advantage is introducing no new components, supporting transactional writes and JOIN queries, and keeping operational complexity low.
At medium scale and in scenarios requiring strongly consistent metadata filtering, many teams choose pgvector to avoid multi-system synchronization issues.
pgvector's default HNSW settings cause insert stalls of more than 2 seconds during batch memory writes.
The HNSW index locks the table during batch inserts, causing write stalls of more than 5 seconds; the solution is to write in batches with INSERT ... ON CONFLICT DO NOTHING and a 50ms debounce window.
In pgvector's HNSW index, the parameter m is the number of neighbor edges kept per node, and ef_construction is the size of the candidate queue during construction; together they determine the recall-latency trade-off at query time.
In pgvector's HNSW index, setting m to 16 and ef_construction to 64 achieves a p99 latency of 12ms under 500 concurrent inserts.
pgvector's index construction consumes considerable memory, and index maintenance cost under high-concurrency writes rises noticeably with data volume, so load testing should be done against actual data scale and concurrency patterns before production deployment.
40 more timeline events
All Facts (20)
The pgvector extension adds vector storage and similarity search capabilities to PostgreSQL, enabling Supabase to serve simultaneously as a traditional database and a vector database.
80%VerifiedPinecone、Chroma、Weaviate是常见的向量数据库产品
80%VerifiedLong-term memory is typically implemented with the help of vector databases (such as Chroma, Pinecone, Weaviate, Milvus, etc.) to enable semantic retrieval.
80%Verified典型的RAG流程包括文档切片、向量化嵌入、存入向量数据库、语义检索、上下文拼接和最终生成六个步骤
75%Verifiedpgvector is an open-source extension for PostgreSQL that natively adds vector data types and similarity search capabilities to the relational database, supporting both exact nearest-neighbor and approximate nearest-neighbor (HNSW, IVFFlat index) retrieval modes.
70%VerifiedPinecone is a fully managed, cloud-native vector database.
70%VerifiedPostgreSQL can support vector retrieval via pgvector.
65%Verified向量存储常用专用向量数据库如Pinecone、Chroma、Weaviate，支持近似最近邻检索
65%VerifiedThe RAG mechanism relies on the approximate nearest neighbor (ANN) search capability of vector databases (such as Pinecone, Weaviate, and Milvus) to achieve millisecond-level retrieval.
65%UnverifiedMaxKB通常集成PostgreSQL的pgvector扩展或独立的向量数据库
75%Unverifiedpgvector was first released in 2021, embedding high-dimensional vector storage and approximate nearest neighbor retrieval capabilities into PostgreSQL.
70%UnverifiedFor applications with vector scale below tens of millions and insensitivity to latency, pgvector's existing indexes may already be sufficient, and there is no need to introduce new dependencies.
60%UnverifiedMainstream vector databases such as Pinecone, Weaviate, Qdrant, and pgvector generally adopt the HNSW approximate nearest neighbor indexing algorithm at their core.
60%UnverifiedFor each query, chatVia.ai runs in parallel a pgvector-based dense vector similarity search and an in-Postgres sparse BM25 keyword search, and merges the rankings using Reciprocal Rank Fusion.
50%Unverifiedpgvector's performance in ultra-large-scale vector scenarios is inferior to that of dedicated vector databases.
50%UnverifiedOntoGr runs on top of FalkorDB or Neo4j, and can optionally use pgvector or Postgres; a background consolidation task handles deduplication, summarization, scoring, and archiving; the LLM and embedder are bring-your-own.
50%UnverifiedDedicated vector databases (such as Pinecone, Qdrant, and Weaviate) offer more extreme performance optimization and distributed scaling for large-scale vector retrieval, while pgvector's advantage is introducing no new components, supporting transactional writes and JOIN queries, and keeping operational complexity low.
50%UnverifiedAt medium scale and in scenarios requiring strongly consistent metadata filtering, many teams choose pgvector to avoid multi-system synchronization issues.
50%UnverifiedThe HNSW index locks the table during batch inserts, causing write stalls of more than 5 seconds; the solution is to write in batches with INSERT ... ON CONFLICT DO NOTHING and a 50ms debounce window.
50%UnverifiedIn pgvector's HNSW index, setting m to 16 and ef_construction to 64 achieves a p99 latency of 12ms under 500 concurrent inserts.
50%