Docs
For agentsStart a data project. Your agent builds it.
Say what you want to study. Your agent sets up LakehouseBox, connects the sources so they keep filling themselves, and answers your questions as the data grows.
Before you start: Claude Code (or any other agent)A question you want to answerTen minutes
Start with what you want to find out
Open Claude Code (or any other agent) and describe your project, not the steps. Point it at these docs.
I want to start an analysis project about air pollution in Madrid, using the city's open data portal (datos.madrid.es). Set up LakehouseBox for it and keep the data coming in, so I can ask questions as it grows. Follow lakehousebox.com/docs/
I've installed the LakehouseBox command line and started the sign-in. One thing for you to do in your browser: open the link, type the code I give you, tick Write and approve.
While you do that, I checked the source: Madrid's open data portal has a real-time air-quality feed, hourly readings for each station and pollutant (NO₂, PM10, PM2.5, O₃…), each flagged valid or not. LakehouseBox can fetch it on a schedule from its own servers, so your computer doesn't need to stay on.
Approve the sign-in
The one step only you can do. Your agent never sees your password; its key is limited to what you approve.
Your turn · in the browser
Approve your agent's access
Open the link your agent gives you. Type the code from your agent's message when the page asks for it (never approve a code someone else sent you), tick Write so it can create tables, and click Approve. No account yet? The same page lets you create one; the free plan is enough.
Not at your computer, or no account yet? Your agent can start without one: it makes a catalog it can write at once, and when you are back it gives you a link and a code to keep it (you sign up or log in there). Unkept, it is deleted after 72 hours. How it works.
Your agent builds the project
It creates the tables, connects the live feed, adds a reference table and leaves notes for the next agent.
Approved with write.
Madrid's air quality now loads into LakehouseBox on its own, and you can start asking questions.
- Live readings: a table gets the city's feed every 20 minutes: 24 stations, every pollutant they measure, only readings the city marks valid. I ran it once by hand: 2,809 readings for today.
- Stations: a second table with each station's name, type (traffic, background, suburban) and coordinates. Every reading matches a station.
- Notes: I saved a short description of the tables in your catalog, so any agent picking this up later reads the data correctly.
Older data: the city also publishes official, checked hourly data for past years. Shall I load it too, so you can compare trends right away?
- A live table that fills itself every 20 minutes, on LakehouseBox's servers
- A stations table to join with
- Notes in the catalog for the next agent
Add the history, then ask
With years of history next to the live feed, real questions have answers.
Yes, load the last 3 years of official data. Then tell me: has NO2 gone down over those years, and which stations improved most?
Yes, NO₂ has gone down clearly, and the drop shows at every station. I loaded the official hourly data from January 2023 to August 2026: 4,008,442 valid readings, the same count as the original files.
| Year | City-wide NO₂ (µg/m³) |
|---|---|
| 2023 | 27.5 |
| 2024 | 23.2 |
| 2025 | 22.4 |
From 2023 to 2025 the yearly average fell about 19%. Most improved: Castellana (traffic) −32%, El Pardo −29%, Plaza Castilla −27%. Plaza Elíptica and Barajas Pueblo improved least and are still the most polluted. Weather varies a lot from year to year, so this is a trend, not proof of what caused it.
Example · verified connection
How does this October compare with October 2023?
At what hours is NO₂ worst near traffic stations?
Email me every Monday a summary of last week's air quality.
Why a lakehouse, and not a spreadsheet
A file answers one question once. A project keeps going: data keeps arriving, history piles up, and you, your agents and your colleagues keep asking new questions over all of it. LakehouseBox keeps that data in tables you own (open Apache Iceberg, stored in Germany), keeps them filling on a schedule with nobody's computer switched on, lets several agents and tools read the same tables, and keeps every version for a while. The free plan includes 5 GB (pricing).
What you can do next
- Bring your files: spreadsheets, CSV and Parquet exports, linked and checked.
- Keep a table filled from a website: weather, prices, open data, on a schedule.
- Use it from your tools: notebooks, Python, DuckDB, Snowflake.
- Share a dataset publicly: anyone can query it, no account needed.
- Work with your team: invite colleagues, give each agent its own limited key.
- Agents that run on a schedule: a weekly summary in your inbox.
- Examples of real projects: Home Assistant, weather stations, a city's open geodata, walked through step by step.
No agent? The account page at lakehousebox.com/app/ creates catalogs, imports files with its Import files button and runs SQL in the browser.