Run embedded
Install ParqDB, register an existing Parquet table, and build an IVF index inside your Python process.
ParqDB documentation
Build and query open vector indexes without moving your data into a proprietary format.
Choose a path
Section titled “Choose a path”Publish for the browser
Publish an immutable IVF-LVQ index to object storage and query it with HTTP Range and WebAssembly.
Understand the system
Learn how ParqDB separates source tables, open indexes, catalog publication, storage, and execution.
Inspect the format
Use the runtime-independent metadata and Parquet schemas to build interoperable tools and engines.
What ParqDB supports today
Section titled “What ParqDB supports today”| Runtime | Storage | Capability | Status |
|---|---|---|---|
| Embedded DataFusion | Parquet | Build and query IVF, IVF-LVQ4, and IVF-LVQ8 | Supported |
| Browser/WASM | Public HTTPS object storage | Query immutable IVF-LVQ indexes over HTTP Range | Experimental |
| Embedded DataFusion | Iceberg | Query exact snapshots through the native provider | Experimental |
| Client/server | Authorized Parquet sources | Build and query through the HTTP API | Experimental |
The embedded DataFusion runtime is the first supported product surface. Review current limitations before planning a production deployment.
Choose a deployment shape
Section titled “Choose a deployment shape”| If you need | Start here | What runs the query |
|---|---|---|
| A local Python application or notebook | Embedded ParqDB | DataFusion in your Python process |
| A public, serverless search experience | Publish for the browser | WebAssembly in each visitor’s browser |
| A shared API for trusted clients | Run a server | The ParqDB process behind HTTP |
All three paths use the same immutable Parquet index format. The browser path has a deliberately smaller feature surface; the limitations page defines the current boundary.
Install
Section titled “Install”python -m pip install parqdbimport parqdb
session = parqdb.connect("./parqdb-data")session.register_parquet("documents", parqdb.datasets.uri("documents"))documents = session.table("documents")Continue with Getting started to build and query your first index.
Before production
Section titled “Before production”- Confirm the source schema and stable key fields in Getting started.
- Tune the IVF index against representative queries and measure recall, latency, and I/O together.
- Set explicit resource and cache limits.
- Review current limitations and the troubleshooting guide.