PostgreSQL is an open-source, ACID-compliant object-relational database management system (ORDBMS) that extends the SQL standard with features such as table inheritance, function overloading, and a rich type system including arrays, JSON, and user-defined types. It supports both relational and document storage paradigms, advanced indexing strategies including GIN, GiST, BRIN, and B-tree, and full-text search, making it a versatile choice for analytical and transactional workloads. Developed from the POSTGRES project at UC Berkeley (1986), it has been maintained as free software since 1996 and is widely regarded as the most standards-compliant open-source relational database.
Content
- PostgreSQL traces its lineage to Michael Stonebraker’s POSTGRES project at UC Berkeley, begun in 1986 as a successor to the INGRES system. POSTGRES introduced an object-relational model with support for user-defined types and rules. In 1996 the project was renamed PostgreSQL to reflect its SQL compliance and released as fully open-source software, bootstrapping a global volunteer community.
- The database engine uses a process-per-connection architecture with multi-version concurrency control (MVCC) that allows readers and writers to proceed without blocking each other. Its query planner is cost-based, supported by extensive statistics, and capable of parallelising scans, joins, and aggregations across multiple CPU cores. Extensions — installable shared libraries — add capabilities ranging from geospatial analysis (PostGIS) and time-series data (TimescaleDB) to vector similarity search (pgvector) and columnar storage (Citus).
- PostgreSQL’s extensibility makes it the preferred persistence layer for AI and data engineering stacks. The pgvector extension, introduced in 2021, allows high-dimensional embeddings to be stored alongside relational metadata and queried with approximate nearest-neighbour indices (IVFFlat, HNSW), collapsing the need for separate vector databases in many applications. This makes PostgreSQL a natural fit for retrieval-augmented generation (RAG) architectures where semantic and structured queries must be composed.
- In 2024-2025, PostgreSQL 16 and 17 delivered improvements to logical replication, parallelism, and JSON path query performance. The pgvector HNSW index, achieving sub-millisecond query latency at billion-scale, has driven explosive adoption in AI application backends. Cloud-managed variants — Amazon Aurora, Google AlloyDB, Neon, and Supabase — are making serverless PostgreSQL accessible at scale, reducing operational burden for teams building AI-native applications.