ParadeDB: product and architecture
ParadeDB in short
ParadeDB upgrades Postgres with a custom index for full-text and vector search, BM25 scoring, filters, facets, aggregations and joins. It is a Postgres extension rather than a fork: you run CREATE EXTENSION pg_search and create a ParadeDB index on your tables, so application data and search live in the same database with standard SQL and ACID transactions.
Buying snapshot
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Fact | Detail |
|---|---|
| Deployment | Drops into any self-managed Postgres as an extension; runs locally with Docker through the install script; ParadeDB Cloud for a managed option. |
| Operating model | A native index on your tables — no sidecar process, no external cluster. One index combines inverted, vector and columnar structures. |
| Query surface | Standard SQL: joins, CTEs, window functions, JSONB, PL/pgSQL, row level security, materialized views, with ACID commits, rollbacks and read-after-write. |
| Licensing | ParadeDB Community is under AGPL-3.0. ParadeDB Enterprise is commercially licensed and included with ParadeDB Cloud; self-hosting Enterprise goes through sales. |
| High availability | Postgres physical replication, plus horizontal read scaling with read replicas, available on the Enterprise plan. |
| Support channels | Slack for questions, GitHub Discussions for longer-form Q&A, GitHub Issues for bugs and feature requests; email covers enterprise support and commercial licensing. |
| Built on | pgrx bridges Postgres and Rust, Tantivy powers full-text and vector search, Apache DataFusion handles OLAP processing. |
Our take
ParadeDB fits teams already on Postgres who want search without a second system to keep in sync, and who are comfortable operating Postgres themselves. The decision splits on node count, not data volume: if a single node with no read replicas carries your search load, Community covers it; if you need high availability, read scaling or a dedicated SLA, you are on an Enterprise quote or on Cloud, which is in private preview.
Where it stands out
- One copy of the data. Search engines and vector stores index a separate copy of your data; ParadeDB keeps application data in one place without schema changes.
- Search behaviour you would otherwise leave Postgres for. Tunable BM25 scoring, the lexical ranking dedicated search engines use, plus typo tolerance, highlighting, phrase, fuzzy, proximity and regex queries.
- Vendor benchmark, narrow scope. P95 text latency of 2.4 ms for ParadeDB 0.26.2 and 5.8 ms for Elasticsearch 8.17, measured single-client on a 28M row Hacker News dataset using rotating two-term title searches with paradedb/benchmarker.
- Customer results published by the vendor. Modern Treasury reports 3x faster writes after one ParadeDB index replaced up to 26 B-tree and GIN indexes on a table; Bilt reports a 95% reduction in query timeouts in resident search with P95 latency down 50%.
- Your Postgres tooling still applies. pg_dump, logical replication and pgBackRest keep working, as do extensions such as pg_partman, pg_cron and PostGIS.
Trade-offs to weigh
- AGPL-3.0 on Community. Clear the license with legal before embedding it in a product you distribute; the commercially licensed Enterprise build is the other route.
- Self-managed means you run the replication. High availability comes from Postgres physical replication on your own infrastructure, so the setup and the on-call stay with your team even once read replica support is licensed with Enterprise.
- The managed option is in private preview. Confirm you can get access to ParadeDB Cloud before planning a migration around it.
- Check your ORM before you commit. Listed integrations cover Drizzle, Django, SQLAlchemy, Rails and EF Core; Prisma sits under "more coming" rather than among them.
- The published numbers are the vendor's own, and of two different kinds. The 2.4 ms and 5.8 ms figures come from one benchmark with a stated dataset, version pair and query shape; the Modern Treasury and Bilt figures are single customer deployments with no such scope stated. Re-run the benchmark on your data, and read the customer figures as examples rather than expectations.
Plans and pricing
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Plan | Price | What it covers |
|---|---|---|
| Self-Managed Community | Fully free forever | A single node with no read replicas, community support; aimed at testing and small projects that don't require multiple nodes. |
| Cloud | Private Preview | Everything in Enterprise, fully managed, one-click deployments — ParadeDB run for you without managing infrastructure. |
| Self-Managed Enterprise | Custom Pricing | Everything in Community plus read replica support, high availability, and dedicated support with an SLA; commercially licensed, arranged with sales. |
What the published plans price is a license and support tier. On both self-managed plans the cost driver is the Postgres infrastructure you run; the Enterprise quote sits on top of that and buys read replica support, high availability and dedicated support with an SLA.