paradedb/paradedb

ParadeDB – Search‑and‑AI inside Postgres

What it is ParadeDB is a PostgreSQL extension that adds a custom index capable of fast full‑text search, BM25 ranking, vector similarity search, hybrid (text + vector) queries, filtering, faceting, aggregations and even joins. Because it lives inside Postgres, your application data and your search/AI index share the same database, the same SQL dialect and the same ACID guarantees.

Key capabilities

Feature	What you can do
Full‑text search	BM25 scoring, top‑K results, term highlighting, custom tokenizers/filters.
Vector search	Store high‑dimensional embeddings (e.g., from LLMs) and retrieve nearest‑neighbors with an efficient index.
Hybrid search	Combine keyword matching and vector similarity in a single query.
Filtering & facets	Apply arbitrary SQL WHERE clauses, compute bucket counts, metrics, and facet counts without leaving the database.
Aggregates & columnar storage	Run OLAP‑style aggregations using Apache DataFusion, with optional columnar storage for speed.
JOINs	Join search results with other relational tables using ordinary SQL joins.

How it works

• The extension is written in Rust and linked into PostgreSQL via the pgrx framework.
• It re‑uses the Tantivy library (a Rust equivalent of Lucene) for both inverted‑index full‑text search and Approximate Nearest Neighbor (ANN) vector search.
• For analytical queries it leverages Apache DataFusion, giving it a modern OLAP execution engine.
• All of this is packaged as a regular PostgreSQL extension, so you install it with CREATE EXTENSION parade and then use standard SQL functions (parade_search, parade_vector_search, etc.) to query.

Installation & deployment

• Quick local start: curl -fsSL https://paradedb.com/install.sh | sh (Docker‑based).
• For production you can self‑host or use ParadeDB Cloud, a fully managed hosted service.
• The project also provides Docker images (paradedb/paradedb).

Ecosystem integrations

• ORMs: adapters for Drizzle, Django, SQLAlchemy, Rails, EF Core (and more planned).
• AI agents: plug‑ins like Agent Skills, MCP Integration, and a Cursor marketplace plugin let LLM‑based agents query ParadeDB directly.
• PaaS: ready‑to‑deploy on Railway, Render, Fly.io, Dokku, DigitalOcean, etc.

Who might use it

• Developers building applications that need both relational data and fast semantic search (e.g., recommendation engines, chat‑bot knowledge bases, e‑commerce product search).
• Teams that want to avoid a separate search stack (Elasticsearch, OpenSearch, etc.) and keep everything under a single PostgreSQL instance.
• Data‑science or analytics teams that want to run vector‑based similarity queries alongside traditional SQL analytics.

Community & support

• Open‑source core under AGPL‑v3, with an optional commercially‑licensed Enterprise edition.
• Active Slack community, GitHub Discussions, issue tracker, and a contribution guide for newcomers.

Bottom line ParadeDB turns PostgreSQL into a unified store for relational data, full‑text search, and vector similarity search, letting you write everything in plain SQL while still getting the performance of dedicated search engines.