An analytics warehouse that answers in milliseconds
A managed columnar warehouse built for BI workloads — sub-second aggregations over billions of rows, open Parquet storage, and zero infrastructure to run.
- Columnar engine tuned for dashboard queries
- Open Parquet storage — your data stays yours
- Queries are never metered — pay for storage, not questions
FROM sales.orders GROUP BY 1
Under the hood
Sub-second answers, open storage
Import a slow source once; the columnar warehouse serves dashboards in milliseconds while your durable copy stays open Parquet.
Serves interactive dashboards in milliseconds.
A durable, portable copy your own engines can read.
Import mode
From four seconds to eighty milliseconds
Point the warehouse at a slow source and import it once; the dashboard that crawled now answers instantly — and every consumer shares one governed surface.
Same dashboard query
~50× faster — p95 under 1s on a 10M-row dataset
One governed surface
SQL engine
RLS, caching, and the AI agent apply automatically — one more governed source.
A closer look
Inside Warehouse
Import mode, milliseconds out
Turn a slow live source or a saved query into a managed columnar dataset that serves dashboards in sub-second time — without running a cluster of your own.
- Columnar engine tuned for aggregation-heavy BI
- Sub-second group-bys over large datasets
- No sizing, vacuuming, or index tuning
Open storage, no lock-in
Every dataset lands as open Parquet on object storage — a durable copy you can point your own engines at. The warehouse copy is just the fast serving layer.
- Open Parquet on object storage you can own
- Readable by your own engines — the inverse of lock-in
- Tiered hot columnar plus lake history, one query surface
Governed, and never metered
Imported data inherits row-level security, result caching, and the AI analyst automatically — it's just one more governed source. And you pay for storage, never for the questions you ask.
- Row-level security and caching applied automatically
- Atomic refresh — old data or new, never a mix
- Queries and dashboard reads are never metered
What you can do with Warehouse
Columnar speed
Aggregations, group-bys, and filters return in milliseconds — dashboards stay interactive even on large datasets.
Import mode
Turn a slow live source or a saved query into a managed dataset that serves dashboards in sub-second time.
Open, portable storage
Every dataset lands as open Parquet on object storage — a durable copy you can point your own engines at. No lock-in.
Atomic refresh
Refreshes swap in atomically, so dashboards see the old data or the new — never a half-loaded mix.
Inherits your governance
Imported data automatically gets row-level security, result caching, and the AI analyst — it's just one more governed source.
Queries never metered
Priced on the storage you use, not the questions you ask. Dashboards and viewers query as much as they like.
Milliseconds
Dashboard query latency
Open Parquet
Portable, no lock-in
Unmetered
Pay for storage, not queries
How it works
Load with Pipelines
Point any connector at the warehouse
Organize datasets
Schemas, tables, and materialized views
Query at speed
Dashboards and SQL both hit the same engine
Scale automatically
Storage and compute grow with your data
Frequently asked questions
No. BI connects directly to 141 sources. The warehouse is for teams that want a managed, fast serving layer without running their own.