Skip to main content

Demo warehouse

Try SignalPilot (/demo-db) gives you a real warehouse and a real dbt project in about a minute, without touching your own data. It is the fastest way to see what the agent does before pointing it at anything that matters.

What it sets up​

  1. Add a demo warehouse — forks a private branch of a seeded Postgres warehouse for you and registers it as a connection. It is your branch: build on it, break it, drop it.
  2. Clone the demo dbt project — the project the warehouse was built from, so the agent has model code to read, not just tables.
  3. Connect your agent via MCP — shows the MCP URL and key to paste into Claude Code or another client.

Remove demo connection cleans up when you are done.

The suggested workflows​

The page ships four prompts worth running in order, because each one shows a different part of the system:

  • Build the warehouse on your branch — the agent runs a dbt build against your own branch, so you see governed writes happening somewhere disposable.
  • Build the revenue-cycle marts on your branch — a larger build, exercising the 8-step dbt workflow and the verifier agents.
  • Investigate a number before you report it — the interesting one. The demo data contains defects that look plausible, so this shows the difference between an agent that trusts a mart and one that checks its grain.
  • Load the knowledge base, then use it — populate the knowledge base and watch the next answer change.

After the demo​

The same shape works on your warehouse: add a connection, import your dbt project, then let the agent work. Two things worth doing early:

  • Write your first knowledge entries about the traps in your data.
  • Set up an eval set with a handful of questions whose answers you already know, so improvements are measurable rather than felt.