ParadeDB — Elasticsearch-Quality Search Inside Postgres

Introduction

ParadeDB extends PostgreSQL with Elasticsearch-grade full-text search and analytics capabilities. Instead of syncing data to a separate search engine, you run BM25 search, sparse vector queries, and columnar analytics directly inside your Postgres database.

What ParadeDB Does

• Adds BM25 full-text search to Postgres via the pg_search extension
• Supports hybrid search combining BM25 text scoring with vector similarity
• Provides columnar storage and vectorized analytics via the pg_analytics extension
• Integrates with standard Postgres tooling (psql, ORMs, connection poolers)
• Eliminates the need for a separate Elasticsearch or OpenSearch deployment

Architecture Overview

ParadeDB ships two core Postgres extensions. pg_search implements an inverted index using the Tantivy search library (Rust-based Lucene equivalent) and exposes BM25 scoring through a custom Postgres operator. pg_analytics adds a columnar table access method built on Apache Arrow and DataFusion for analytical queries. Both extensions run inside the Postgres process and use standard Postgres transactions.

Self-Hosting & Configuration

• Run the ParadeDB Docker image, which is standard Postgres with extensions pre-installed
• Alternatively, install pg_search and pg_analytics as extensions on an existing Postgres instance
• Create BM25 indexes with standard CREATE INDEX syntax using the bm25 access method
• Configure tokenizers, analyzers, and search options per index
• Compatible with managed Postgres providers that support custom extensions

Key Features

• BM25 full-text search with relevance scoring, phrase matching, and faceted search
• Hybrid search combining keyword (BM25) and semantic (vector) results in one query
• Sparse vector search for learned sparse retrieval models like SPLADE
• Columnar analytics with vectorized execution for OLAP-style queries
• Standard SQL interface with no proprietary query language to learn

Comparison with Similar Tools

• Elasticsearch — Purpose-built search engine with its own query DSL and operational complexity; ParadeDB runs inside Postgres
• OpenSearch — AWS fork of Elasticsearch with similar architecture; still requires a separate cluster
• Meilisearch — Fast typo-tolerant search API, but standalone service and not SQL-based
• Typesense — Developer-friendly search engine, but requires data sync from your primary database
• pgvector — Postgres extension for vector similarity only; ParadeDB adds BM25 text search and hybrid scoring

FAQ

Q: Can I use ParadeDB with my existing PostgreSQL database? A: Yes. pg_search and pg_analytics are standard Postgres extensions that can be installed on any Postgres 14+ instance.

Q: Does ParadeDB replace Elasticsearch for all use cases? A: ParadeDB covers full-text search, hybrid search, and basic analytics. For specialized use cases like log aggregation at massive scale, a dedicated search engine may still be appropriate.

Q: How does search performance compare to Elasticsearch? A: ParadeDB uses Tantivy, a Rust-based search library that benchmarks competitively with Lucene. For most application search workloads, performance is comparable.

Q: What is the license? A: ParadeDB is licensed under AGPL-3.0.

Sources

• https://github.com/paradedb/paradedb
• https://docs.paradedb.com