ParadeDB Docs

ParadeDB is a search engine built into Postgres. Its pg_search extension brings BM25 relevance ranking, faceting, hybrid search, and aggregations to your tables in standard SQL. Everything runs inside your system of record with Postgres’ transactional guarantees, not in a second system you have to keep in sync.

ParadeDB Cloud is coming soon. We’re building a fully managed ParadeDB: first-class search from Postgres, with a developer experience to match. Join the waitlist.

Who is ParadeDB for?

You’re likely a good fit for ParadeDB if any of the following sound like you:

Postgres is your primary database (managed or self-managed) and you’d rather build on it than around it.
You’ve outgrown Postgres’ built-in search (tsvector and the GIN index) and hit performance bottlenecks or missing features like BM25 scoring, fuzzy matching, or faceting.
You’re evaluating a search engine like Elasticsearch, but don’t want to run (and continuously sync) a second system alongside your database.

Why ParadeDB?

For teams that already use Postgres, ParadeDB is the simplest path to one Postgres for your application data, full-text search, vector retrieval, and aggregations.

Zero ETL Required

Syncing Postgres with an external search engine like Elastic is a time-consuming, error-prone process that involves babysitting ETL pipelines and debugging data inconsistencies. ParadeDB eliminates this class of problems because search, vectors, and aggregations live right next to your data. pg_search is a pure Postgres extension: no fork, no separate server. Install it in your primary Postgres and your search index stays current automatically: index updates happen in the same transaction as your writes, with no pipeline to maintain. If you can’t install extensions on your primary, see other deployment options.

Search That Feels Like Postgres

In ParadeDB, a search query is just SQL. You use the operators and functions you already know, with full support for JOINs, so there’s no need to denormalize your existing schema.

-- Index a table for full-text search
CREATE INDEX search_idx ON mock_items
USING bm25 (id, description, rating)
WITH (key_field='id');

-- Rank by relevance, then filter and sort like any other query
SELECT description, pdb.score(id)
FROM mock_items
WHERE description ||| 'running shoes' AND rating > 2
ORDER BY score DESC
LIMIT 5;

ParadeDB accelerates standard SQL queries. That includes:

Text: relevance-ranked full-text queries like the one above.
Aggregations: the same index serves faceted search and aggregations, pushing filters and aggregates directly into the index instead of computing them afterward.
Vectors: pair ParadeDB with pgvector to combine full-text and semantic search in one query, with no separate vector database to keep in sync. A high-performance, drop-in compatible version is coming to the ParadeDB index soon.

One Index Behind Every Query

Behind the SQL is a single custom index: the BM25 index, built on Tantivy, the Rust port of Lucene. It goes toe-to-toe with dedicated search engines on full-text performance, often coming out on top. See how it’s built →

As Reliable As Postgres

ParadeDB supports Postgres transactions and ACID guarantees. Data is searchable immediately after it’s written, and durable thanks to Postgres write-ahead logging. See Guarantees for the details, including isolation levels and replication safety.

ParadeDB vs. Alternatives

People usually compare ParadeDB to two other types of systems: OLTP databases like vanilla Postgres and search engines like Elastic.

	OLTP database	Search engine	ParadeDB
Primary role	System of record	Search and retrieval engine	System of record and search engine
Examples	Postgres, MySQL	Elasticsearch, OpenSearch
Search features	Basic FTS (no BM25, weak ranking)	Rich search features (BM25, fuzzy matching, faceting, hybrid search)	Rich search features (BM25, fuzzy matching, faceting, hybrid search)
Analytics features	Not an analytical DB (no column store, batch processing, etc.)	Column store, batch processing, parallelization via sharding	Column store, batch processing, parallelization via Postgres parallel workers
Lag	None in a single cluster	At least network, ETL transformation, and indexing time	None in a single cluster
Operational complexity	Simple (single datastore)	Complex (ETL pipelines, managing multiple systems)	Simple (single datastore)
Scalability	Vertical scaling in a single node, horizontal scaling through Kubernetes	Horizontal scaling through sharding	Vertical scaling in a single node, horizontal scaling through Kubernetes
Language	SQL	Custom DSL	Standard SQL with custom search operators
ACID guarantees	Full ACID compliance, read-after-write guarantees	No transactions, atomic only per-document, eventual consistency, durability not guaranteed until flush	Full ACID compliance, read-after-write guarantees
Update & delete support	Built for fast-changing data	Struggles with updates/deletes	Built for fast-changing data

Production Readiness

ParadeDB launched out of the Y Combinator (YC) S23 batch and has run in production since December 2023. ParadeDB Community, the open-source version of ParadeDB, has been deployed over 1.5 million times via its Docker image, and the pg_search extension has been installed over 250,000 times. ParadeDB Enterprise, the durable and production-hardened edition of ParadeDB, powers core search and analytics use cases at enterprises ranging from Fortune 500s to fast-growing startups. A few examples include:

Alibaba Cloud, the largest Asia-Pacific cloud provider, uses ParadeDB to power search inside their data warehouse. Case study available.
Bilt Rewards, a rent payments technology company that processed over $36B in payments in 2024. Case study available.
Modern Treasury, a financial technology company that automates the full cycle of money movement. Case study available.
Span¹, one of the fastest-growing AI developer productivity platforms.
TCDI¹, a giant in the legal software and litigation management space.

1. Case study coming soon.

Next Steps

You’re now ready to jump into our guides.

Getting Started

Get started with ParadeDB in under five minutes.

Architecture

Learn how ParadeDB is built.

Reference

API reference for full text search and analytics.

Deploy

Deploy ParadeDB as a Postgres extension or standalone database.

Documentation

Search without a second system

Who is ParadeDB for?

Why ParadeDB?

Zero ETL Required

Search That Feels Like Postgres

One Index Behind Every Query

As Reliable As Postgres

ParadeDB vs. Alternatives

Production Readiness

Next Steps

Getting Started

Architecture

Reference

Deploy

​Who is ParadeDB for?

​Why ParadeDB?

​Zero ETL Required

​Search That Feels Like Postgres

​One Index Behind Every Query

​As Reliable As Postgres

​ParadeDB vs. Alternatives

​Production Readiness

​Next Steps

Getting Started

Architecture

Reference

Deploy

Who is ParadeDB for?

Why ParadeDB?

Zero ETL Required

Search That Feels Like Postgres

One Index Behind Every Query

As Reliable As Postgres

ParadeDB vs. Alternatives

Production Readiness

Next Steps