Home
-
Browser worksheets
A Monaco SQL editor with tabs, a results grid, and CSV export. Write and run queries together without emailing
.sqlsnippets around. -
Shared Iceberg catalog
Browse schemas and tables with sample rows. Every table is Iceberg-native, with snapshot history and "query at this snapshot" time travel.
-
Governed workspaces
Per-workspace permissions scoped down to catalog/schema/table grants, service-account machine auth, Apache Polaris catalog integration, and a full audit trail of who ran what, where, and when.
-
Transparent compute
You pick the DuckDB agent per query. No opaque optimizer, no surprise costs, no hidden resource allocation. Scale by adding agents.
-
Right-sized memory
Each query's memory reservation is estimated from DuckDB's
EXPLAINplan, so cheap queries pack in and heavy ones queue instead of OOM-ing the agent. -
Query profiles
After each run, inspect an interactive operator graph with rows, bytes, and timing per step, plus flags for spills, scan blow-ups, and bad estimates.
-
AI data assistant
An opt-in, governed chat assistant that browses the catalog, writes and runs SQL, and proposes worksheet edits — audited like any other user.
Explore the docs¶
-
Quickstart
Stand up the full stack with one
docker composecommand and run your first query. -
Architecture
The control-plane / compute split, invariants, and data flow — the defining idea behind DuckHaven.
-
Deployment
Install, add agents, front the stack with TLS, update, and back up your data.
-
Administration
Day-2 operations: query queueing and concurrency, agent management, and the operator runbook.
-
Configuration
Every control-plane and agent environment variable, consolidated into one reference.
-
Contributing
Local development, the four-layer test architecture, the design system, and cutting a release.