Claude
ChatGPT01
Ingest
- sales.csv
- events.parquet
- https://…
- And more…
Upload files, refresh a table from a URL on a schedule, or receive rows from devices and applications.
Ways to bring data →For AI builders and data teams
Data infrastructure
for AI agents.
Give your agents a place to work with data.
And keep what they build.
LakehouseBox stores and maintains your analytical tables, so agents can query, save results and continue the work across tools and sessions.
5 GB free · No credit card · Managed in Germany
In Claude Code, add the LakehouseBox plugin (skills and connector):
/plugin marketplace add lakehousebox/agent-skills /plugin install lakehousebox@lakehousebox
Then ask: “set up a lakehouse and load sales.csv”.
In a terminal, add the LakehouseBox plugin to Codex (skills and connector):
codex plugin marketplace add lakehousebox/agent-skills codex plugin add lakehousebox@lakehousebox
Then ask: “set up a lakehouse and load sales.csv”.
Claude Code v2.1.220
Opus 5.5 with high effort · Claude Team
~/sales
set up a lakehouse for our sales and load sales.csv
Thinking…
I’ll create a catalog in LakehouseBox and import the file.
Bash(lhbox catalog create sales)
✓ catalog sales ready · Frankfurt, Germany
Bash(lhbox table import sales.retail.orders sales.csv)
✓ 184,302 rows · Apache Iceberg · one commit
Worked for 23s
why did sales drop last week?
Querying…
I’ll query the table with DuckDB on this machine.
Bash(lhbox duckdb --catalog sales -- -c "…")
SELECT date_trunc('week', day) AS week, sum(amount) AS sales FROM sales.retail.orders GROUP BY 1 ORDER BY 1 DESC LIMIT 4;
| week | sales |
|---|---|
| 2026-10-12 | 142,320 |
| 2026-10-05 | 151,402 |
| 2026-09-28 | 150,991 |
| 2026-09-21 | 149,338 |
4 rows · 0.3 s
Called lhbox duckdb 2 more times
Found it: demand didn’t fall, card payments did.
– north card orders fell to near zero from Tue 14:00
– cash orders and basket size held flat everywhere
– the week ends 6% down, all of it Tue–Wed
ran in DuckDB on this machine · tables in LakehouseBox
Worked for 18s
save the daily breakdown for the team
Saving…
Saved retail.weekly_drop (14 rows). Anyone on the team can query it from their own assistant.
❯
⏵⏵ auto mode on (shift+tab to cycle) · esc to interrupt
ChatGPT · LakehouseBox connector
the sales manager, a few minutes later
Is the north back to normal? Use weekly_drop.
Thinking
Called LakehouseBox › query (2)
Yes. Card orders in the north were back to their usual level by Thursday 09:00, and Friday finished 2% above the week before.
Which stores were hit hardest?
Thinking
Called LakehouseBox › query
Bilbao Centro and Oviedo: about 70% of the lost card revenue between them.
Same table the data lead saved in Claude Code. No copy, no export.
DuckDB 1.5.5+
PyIceberg 0.12
Spark 3.5
Trino 483
Polars 1.44
A foundation for the next step
Ingest, store, analyze
and run AI analytics loops.
Bring data in many ways, store it in Apache Iceberg catalogs, analyze it with DuckDB or your own engine, and let AI agents close the loop. We handle the rest.
02
Store
- sales.orders
- marketing.events
- product.catalog
- support.tickets
- New table
Apache IcebergCatalog sales
Apache Iceberg catalogs, with storage, per-table credentials and table maintenance included. Your data stays portable.
How it works →03
Analyze
SELECT category, sum(amount) AS sales FROM sales.orders WHERE day >= current_date - 30 GROUP BY 1 ORDER BY sales DESC;
| category | sales |
|---|---|
| Hardware | 482,291 |
| Software | 320,118 |
| Services | 116,992 |
| Accessories | 87,421 |
Query the same tables with DuckDB, Spark, Trino, Polars or PyIceberg, or from a chat through MCP.
Choose your engine →04
Close the loop
An agent’s saved results
| run | rows | saved |
|---|---|---|
| run_1019 | 5,231 | 10-19 |
| run_1012 | 12,446 | 10-12 |
| run_1005 | 8,904 | 10-05 |
| run_0928 | 15,332 | 09-28 |
Results saved to sales.analysis on every run
Agents save their results as tables and pick them up in the next session, or run on a schedule.
Agents on a schedule →LakehouseBox Enterprise
Ready for your
organisation.
Connect your agents to company data with corporate sign-in, granular permissions and a record of activity. Use our managed service in Europe or deploy on your own servers.
Managed in Europe or self-hosted · Enabled per organisation
Organisation accessPeopleGroupsActivitySettings
| Principal | Resource | Access | Via |
|---|---|---|---|
| Analytics team | sales | Read & write | Directory group |
| reporting-agent | sales.analysis | Read | Service account |
| data-science | marketing | Read & write | Directory group |
| finance-lead | sales.orders | Read | Table grant |
Audit log · grant added for reporting-agent · 12 s ago
Corporate sign-in
Connect your identity provider and provision people and groups from your directory.
Scoped access
Read or write on a catalog, a namespace or a single table, for people, groups and service accounts.
Audit trail
Review sign-ins, access changes and data changes in your organisation’s audit log.
Capabilities
Built for AI workflows.
Serverless
No servers or clusters to run. Your agent brings the compute.
Many ways to ingest
Files, scheduled URLs, devices and applications.
Automatic table management
Compaction and table maintenance included, with automatic spatial clustering for geometry tables.
Open table formats
Apache Iceberg, readable by any compatible engine.
Bring your own AI
Claude, ChatGPT, Cursor or any MCP host or agent.
Public data over a CDN
Attach open catalogs with no account, served from the edge.
Public data
Public data,
ready to join.
Browse open catalogs or publish your own.
Eighteen datasets live today.
Explore public data →5 GB on publish · Up to 50 GB for reviewed datasets
- places.places
name category Museo Nacional del Prado art_museum Musée du Louvre art_museum Rijksmuseum museum Pergamonmuseum museum - environment.air_quality_daily
date pollutant mean 2026-10-17 PM10 32.4 2026-10-17 PM10 21.6 2026-10-17 NO2 20.1 2026-10-17 PM10 97.5 - weather.daily
day tmax_c tmin_c 2026-10-12 23.9 14.5 2026-10-12 22.2 15.9 2026-10-11 17.2 9.7
Pricing
Simple storage pricing.
Public catalogs Live
For open datasets anyone can query and join with their own.
Up to 50 GB
Free: 5 GB as soon as you publish, up to 50 GB for datasets we review, apart from your private 5 GB.
Publish a catalogPay as you go
For private storage above the free 5 GB.
€0.10 per GB-month
Nothing else is metered. Need more than 5 GB? Write to us and we raise your limit, free.
Full pricingNo request, ingress or compute fees. Egress under fair use. Maintenance included. Full pricing → · Enterprise → · Limits →