Question
Start where the work happens
A person or agent asks through a browser workspace, application, BI tool, notebook, command line, or MCP client.
- Apps & BI
- Notebooks & CLI
- Agents & MCP
The data farm
The Query.Farm platform
A real farm is not one tool β it's fields, barns, orchards, paths, and storage working toward one outcome. Query.Farm works the same way for data access: each component does a job, and together they let teams query scattered data with SQL.
The product map
One system, from question to capability
Follow the six steps from the question someone asks to the data or compute that answers it. Product status is shown at the point where each component fits. Read the map from top to bottom. People and agents ask questions through the tools they already use; query engines plan the work; extensions and VGI add capability; optional policy controls what may be seen; and workers reach the data or compute that answers. DuckDB works today, while VGI opens the path to more query engines over time.
-
Entry point
-
Available today
Query engine
Plan the work with SQL
DuckDB works today. Haybarn packages it for teams that need signed artifacts, curated extensions, and independent releases.
- DuckDB today
- More engines through VGI over time
-
Available today
Integration
Add the missing capability
Load a native extension or use VGI when the work belongs in another process, language, service, or machine.
-
Optional Β· Preview
Policy
Govern access when required
Rowfence can sit in front of a VGI worker and decide which rows, columns, and functions each caller may use.
-
Self-run or shared
Worker
Run it yourself or reuse it
Workers execute close to the capability. Build and operate your own, or discover reusable connectors through Orchard.
-
The outcome
Answer
Reach the data or compute
The worker reaches the API, service, database, file, model, or library and returns Arrow results to the query engine.
- APIs & services
- Databases & files
- Models & custom code
View the complete architecture diagram
Swipe horizontally to inspect the original desktop diagram.
The data farm
Available
Available
Preview
Available
Available
The components, one by one
Each component does one job β connect, run, govern, package, browse, or reuse. Mix the ones you need today; the rest are ready when you grow into them.
Available
Installable SQL capabilities
Extensions
Connect systems and add SQL capabilities.
Install connectors, transformations, sketches, scripting, geospatial, JSON, text, and operational tools without building native extension plumbing yourself.
Preview
Connector marketplace
Orchard
Discover, subscribe to, and reuse connectors.
The marketplace for VGI workers and connectors β find a connector, subscribe, and attach it from DuckDB or Haybarn with one line of SQL.
Available
Vector Gateway Interface
VGI
Run custom or reusable workers from SQL.
Make APIs, services, models, and business logic available as SQL-accessible tables and functions through Apache Arrowβbased workers β in Python, TypeScript, Go, Rust, or Java.
Preview
Row & column security for VGI
Rowfence
Enforce row, column, and function access.
A Cedar-policy proxy that sits in front of any VGI worker and decides, on every query, which rows, columns, and functions a caller can touch β hosted or on your own infrastructure.
Available
Independent distribution
Haybarn
Bring DuckDB into the enterprise.
Same SQL and database files as DuckDB, with faster object-store reads, signed artifacts, build provenance, curated extension defaults, and mirrorable npm / PyPI channels.
Available
Web workspace for VGI
Cupola
Browse and query workers in the browser.
An open-source, WASM-powered web workspace for VGI: browse catalogs, run SQL, pivot data, and use a built-in AI agent β free and available now.
All together
One data platform
These components work as one platform for building, operating, and reusing SQL-accessible connections β adopt them piece by piece. Extensions add capability, VGI runs the workers, Rowfence governs who sees what, the Orchard shares them, Haybarn ships the engine, and Cupola is where you sit down and query.
229M+
Extension loads per year
Available today
Ship with what's ready now
Three production-ready components you can adopt immediately β independently or together.
DuckDB Extensions
Installable SQL capabilities
Specialized extensions for connectors, transformations, sketches, scripting, and more.
- Connectors, transformations, sketches, and scripting
- All open source
- Production-ready today
VGI
Vector Gateway Interface
Extend DuckDB with functionality in Python, TypeScript, Go, Rust, or Java. Apache Arrow is the cross-language glue β no FFI, no rebuilds, no version locking.
- DuckDB-side extension + Python function framework
- SDKs for Python, TypeScript, Go, Rust, and Java
- Apache Arrow IPC over pipes, sockets, or HTTP
Haybarn
An independent distribution of DuckDB
The DuckDB source, rebuilt as branded binaries with an independent release cadence and a signed extension ecosystem. ABI- and file-format-compatible with upstream DuckDB.
- Independent release cadence and versioning
- Signed extensions, SLSA build provenance
- Install via npx, uvx, pip, or direct download
Need something custom?
We build DuckDB extensions for clients who need specific functionality β from connectors to compute engines.