Build It in One Prompt

You have a project and a question. This page hands both to Kai in a single prompt and lets it build the entire pipeline — the loading, the SQL, the daily schedule, and a map app — while you watch. Step 2 of the Getting Started arc.

What you need

• A project — Get a Project gets you one.
• Nothing else. No installs, no credentials, no files to download.

The prompt

Open Kai Agent in the top bar and paste this:

Load the four CSVs from https://help.keboola.com/getting-started/ — occurrences.csv,
species.csv, depth_zones.csv and basins.csv — into Storage as a single HTTP data source
configuration called "Octopus sightings". Then join them into one table called
octopus_atlas that has, for every sighting, its species display name, its ocean basin and its
depth zone. Wire the loading and the joining into a flow called "Octopus pipeline"
that runs every day at 6am UTC. Finally build me a data app called "Octopus map" — a
world map of every sighting coloured by depth zone, with a field where I type my own latitude
and longitude and it tells me the nearest recorded sighting.

That is one sentence per thing you want, in the order you want it. Nothing in it names a component, a bucket, or a SQL dialect — Kai picks those.

What happens next

Kai works for ten to fifteen minutes and asks you to approve each change it makes to the project. In our run it asked eleven times. The dialog offers Approve, Decline and Always allow; approving one at a time is the point the first time round, because each dialog shows you the exact configuration it is about to write.

It moves through the request in order:

1. Reads the files first. Before configuring anything it fetches the CSVs and looks at their headers, so the columns it joins on are the real ones.
2. Builds the data source. One HTTP configuration with four rows — one per file — and a skip-lines processor, because the files carry a header row.
3. Writes the SQL. A Snowflake transformation joining the sightings to all three lookup tables, with the output table declared in its output mapping.
4. Assembles the flow. Loading in the first phase, joining in the second, so the tables have landed before the SQL runs — plus the schedule you asked for.
5. Writes an actual app. Not a template: it edits package.json, writes the frontend, installs the dependencies, type-checks the code, commits it to the app’s git branch, waits for the container, and only then shows you the live preview.

Watch the plan, not the keystrokes

Kai states what it is about to do before it does it. If a step looks wrong — the wrong table, the wrong schedule — Decline and say what you wanted instead. It is a conversation, not a batch job.

What it cannot do for you

One thing in this guide is yours alone: authorizing Google. If you also want the table delivered to a spreadsheet, Kai can create and configure the Google Sheets destination, but the consent screen lives in your own Google account, so you sign in yourself and hand the conversation back. That is why the prompt above stops at the app — see Deliver the Answer for the spreadsheet step.

Check it worked

Kai finishes with a summary of what it made. Verify it yourself — the next page, What Kai Built, walks every artifact and the numbers each one should show.

The short version: octopus_atlas should hold 10,000 rows — the same count as the raw sightings table, which is the quickest proof that the joins matched every row without duplicating any.

Next: What Kai built →