How to Self-Host ParadeDB in PostgreSQL

How do I self-host ParadeDB inside PostgreSQL?

Self-hosting ParadeDB lets you run vector search inside your own PostgreSQL instance for full control and lower latency.

Welcome to the Galaxy, Guardian!
You'll be receiving a confirmation email

Follow us on twitter :)

Oops! Something went wrong while submitting the form.

Description

Example H2

Example H3

Why choose self-hosting over ParadeDB Cloud?

Self-hosting keeps embeddings and user data inside your VPC, avoids SaaS fees, and lets you tune PostgreSQL exactly for your workload.

Which files do I need to run?

ParadeDB ships a single Docker image that bundles PostgreSQL, the vectors extension, and ParadeDB’s SQL procedures. No extra build steps.

How do I spin up ParadeDB with Docker Compose?

Create docker-compose.yml and run docker compose up -d. ParadeDB starts on port 5432 with persistent volume mapping for /var/lib/postgresql/data.

services: paradedb: image: paradedb/paradedb:latest ports: - "5432:5432" environment: POSTGRES_PASSWORD: secret volumes: - pgdata:/var/lib/postgresql/data volumes: pgdata:

Need SSL?

Mount your certs and set POSTGRES_SSL=on. ParadeDB uses the standard PostgreSQL parameters.

Can I install ParadeDB in an existing cluster?

Yes. Copy paradedb--*.sql files to $PGDATA/extension or use CREATE EXTENSION if the package is in share/extension.

What is the minimal SQL to enable ParadeDB?

Connect as superuser, then:

CREATE EXTENSION IF NOT EXISTS vectors; CREATE EXTENSION IF NOT EXISTS paradedb;

How do I store product embeddings?

Add a vector column to products and create an ivfflat index.

ALTER TABLE products ADD COLUMN embedding vector(384); CREATE INDEX idx_products_embedding ON products USING ivfflat (embedding vector_l2_ops) WITH (lists = 100);

How do I query for similar products?

Use ParadeDB’s helper to embed the search phrase, then sort by L2 distance.

SELECT id, name FROM products ORDER BY embedding <=> paradedb_embed('lightweight laptop') LIMIT 5;

How do I back up the instance?

Because ParadeDB is plain PostgreSQL, use pg_dump, pg_basebackup, or any managed-storage snapshot strategy.

Best practice: tune work_mem & shared_buffers

Set shared_buffers to 25% RAM and work_mem to at least 32MB to avoid temporary files during ANN queries.

Common mistakes and fixes

Missing ivfflat analyze: run VACUUM ANALYZE products; after bulk inserts so the planner uses the index.

Too few lists: a low lists setting slows recall. Reindex with 100-200 lists for >1M vectors.

Why How to Self-Host ParadeDB in PostgreSQL is important

How to Self-Host ParadeDB in PostgreSQL Example Usage


-- Find customers who bought products similar to a search phrase
WITH similar_products AS (
  SELECT id
  FROM products
  ORDER BY embedding <=> paradedb_embed('gaming laptop')
  LIMIT 20
)
SELECT c.id, c.name, SUM(o.total_amount) AS total_spent
FROM customers c
JOIN orders o  ON o.customer_id = c.id
JOIN orderitems oi ON oi.order_id   = o.id
WHERE oi.product_id IN (SELECT id FROM similar_products)
GROUP BY c.id, c.name
ORDER BY total_spent DESC
LIMIT 5;

How to Self-Host ParadeDB in PostgreSQL Syntax


-- Enable extension in a fresh ParadeDB container
CREATE EXTENSION IF NOT EXISTS vectors;
CREATE EXTENSION IF NOT EXISTS paradedb;

-- Embed products and index
ALTER TABLE products ADD COLUMN embedding vector(384);
CREATE INDEX idx_products_embedding ON products USING ivfflat (
    embedding vector_l2_ops
) WITH (lists = 100);

-- Query nearest neighbours with ParadeDB helper
SELECT id, name, price
FROM products
ORDER BY embedding <=> paradedb_embed('noise-cancelling headphones')
LIMIT 10;

Common Mistakes

Forgetting to run VACUUM ANALYZE after bulk loading embeddings. Without updated statistics, the planner may ignore the ivfflat index and force a full scan.
Setting lists=10 for millions of vectors. Low list counts hurt recall and increase latency. Recreate the index with a higher lists value (100-200).