Analytical storage for agents
Your agents bring the compute. We take care of the storage.
Apache Iceberg catalogs and storage for the data your agents create and use. Run analysis where your agent runs, and keep the results ready for the next session.
5 GB free · No credit card · Hosted in Europe
Want to run it yourself? Explore self-hosting →
› SELECT bed, avg(moisture)
FROM garden.moisture
GROUP BY bed;
Illustrative query, not a recorded run.
Paste the prompt into your agent. You approve the connection in your browser. Your agent receives its own revocable access.
DuckDB
PyIceberg
SparkSee it in action
Ask a question. Keep the answer.
01Ask a question
Not the rain. 1.2 mm fell on Friday morning and nothing since. Beds 2 and 3 were watered twice, Saturday 04:00 and Sunday 08:00, and drained back within the day each time. Bed 1 got water once, Saturday around 17:00, and has held 21 % since. Bed 4 climbed slowly all weekend, 25 % to 32 %.
Your agent starts with data in your catalog.
02Analyse with DuckDB
The query runs in your agent’s environment.
03Save the result
CREATE TABLE garden.bed_moisture_hourly AS SELECT … ;
A new Iceberg table is ready for another session or an authorised teammate.
The session ends. The data stays.
How this ran
| Data | garden.sensor_readings, 185,884 rows, a real garden's sensors, stored in a LakehouseBox catalog |
| Query | DuckDB 1.5.5 on a laptop, in the agent's environment · 363 ms |
| Saved | back to the catalog in 1,231 ms · reopened by a new process 340 ms after attaching |
| When | Recorded 2026-09-20 against the live service. The chart is drawn from the saved table. The CREATE TABLE ran in DuckDB; the statement above is abbreviated. |
A shared home for analytical data.
Keep your work
Tables persist across sessions, so your next analysis starts where the last one finished.
Share access
Give people and agents access to the catalogs they need, and revoke it when they no longer do.
Choose your compute
Work with supported engines on your laptop, server or agent environment.
Keep tables maintained
On the hosted service, small files are merged, old versions pruned and orphan files removed for you.
Public data
Public data deserves a home, too.
Give useful datasets a place where people and agents can discover and query them. Publish in open formats, with the source, licence and context others need to use them.
Examples of what could live here, not datasets that exist yet. See the catalog preview →
Free storage for public data
We’re preparing free hosting for public datasets: open tables anyone can JOIN with their own. Interested in publishing one? Get in touch.
Our cloud. Your infrastructure. Same LakehouseBox.
Hosted in Europe
Start with our hosted service. We operate the storage, the catalog and the table maintenance on Hetzner servers in Nuremberg, Germany, as a Spanish company. No US-controlled company operates any part of the service or holds your data. Whether a law such as the US CLOUD Act can reach a provider depends on that provider’s ties to the United States; we list every provider in our path, and every dependency outside the EU, on the sovereignty page.
Operator, location, applicable law and the dependency ledger, stated as facts: the sovereignty page.
Start freeSelf-hosted beta
Run LakehouseBox on your own infrastructure with the same agent-facing tools and engine connections. It is a handful of containers, one compose file and one state directory, so it runs wherever Docker runs. Today that is the compose file and the deploy scripts, not a packaged installer; licence and support terms are being decided.
Run today on Hetzner; the other providers are untested. Anything with Docker and a disk should do.
Explore self-hosting →The LakehouseBox Journal
Does your agent need a data warehouse?
Before buying more infrastructure, try the compute you already have. An agent, DuckDB and shared Iceberg tables can be enough to get started, and they show you what your workload actually needs.
Read the story“Start with the compute you already have.”
Pricing
Start with a place for your data.
- Apache Iceberg catalog and storageincluded
- Access for people and agentsincluded
- Managed table maintenanceincluded
- Connections for supported enginesincluded
- Up to 3 catalogs per project; 20 snapshots or 7 days of history per tableenforced
- Guidelines, not enforced today: 50 tables and 10 namespaces per catalog, 50,000 objectsadvisory
Query compute runs in your environment. LakehouseBox does not bill you for query execution.
Beta. Not yet, and we say so
- One site, no SLA. Off-host copies every 5 minutes for accounts and grants, hourly for table metadata, nightly for data files; restore rehearsed in about 35 minutes; no automatic failover yet. High availability is planned.beta
- Paid plans wait for the company registration; the payment provider will be European.later
- Public datasets. Designed and measured, not switched on.coming
- Snowflake. The catalog integration connects; the data path needs work on both sides.limited
- A full web dashboard. Your account page is small: catalogs, a SQL explorer, access tokens, members, usage.partial
Limits, backups and what is measured: the docs.