Before you begin
- A project to work in (create one from Projects → New project).
- At least one connected integration if your pipeline starts from an external API. See Integrations.
Steps
1
Open the AI Architect
From your project, start a new chat. The Architect splits the screen: chat on the left, workflow graph on the right.
2
Describe what you want
Type what the pipeline should do in plain English—for example, “Pull new Stripe invoices every hour, flag the ones above $10k, and write them to a dashboard.”The Architect proposes a step DAG with the right step types and pre-coded operators wired together.
3
Iterate on the graph
Review the proposed steps. Ask the Architect to add, remove, or reconfigure steps. Each step in the graph shows its type, operator, and parents.See Step types for what each role does and where it can sit in the DAG.
4
Deploy
When the graph is right, deploy the project. Deploy creates the runtime resources (databases, dashboards) and makes the pipeline runnable.
5
Run it
Trigger the pipeline with Run all from the project page, or attach a schedule to a database step so it refreshes on a cron. See Runs.
Next steps
- Understand the step types and what they’re allowed to do.
- Pick the right operators for ingest, transform, AI, and storage.
- Configure triggers for scheduled and cascade refreshes.