Skip to main content
A worked example: a team training dashboard — completion rates, a couple of charts, filterable by department and office over a date range. Connect in Builder mode first.

1. Describe it

You don’t need function names, SQL, or the SDK — your agent works those out.

2. Review the functions it creates

Your agent reuses what exists and creates what’s missing. When it shows you a result, confirm it:
It ends up with functions like: _COMPLETION_BY_DEPARTMENT has no department filter — you don’t filter by the field you’re breaking down by. Each breakdown is its own function.

3. Let it build the app

Your agent generates the dashboard and wires your filter controls to the function parameters through the SDK. The published app authenticates with its own API key — set it as your platform’s app secret, or the dashboard loads blank.

4. Review and iterate

Adjust in plain language:
Check that filters change the numbers, that clearing an optional filter shows everything, and that the charts read from the right fields.

5. Ship it

Ask your agent to confirm its pre-ship check: functions reused, runtime SDK-only, loading and error states present. Once it passes, each function call returns quickly with no AI in the loop.