What you are about to build

The tour shows you the platform one feature at a time. This track does the opposite: it builds one real app and turns each feature on as the app needs it.

The app is Lumen, an incident desk — a list of incidents, a database behind it, an endpoint that summarises an incident with AI, a workflow, and a pipeline that runs overnight. You start from an empty directory and finish with all of it, but the stages are cumulative rather than all-or-nothing: you may stop after any one and still have something real running.

The six stages

• Stage 1 — Get it live. Install the template and deploy it. A URL, a handbook, a code review, and an app your colleagues can see.
• Stage 2 — Give it data. A managed Postgres database, bound to the app and seeded on boot.
• Stage 3 — Give it AI. An endpoint that summarises an incident through the Dibbla AI gateway, authenticated with your own token.
• Stage 4 — Give it a workflow. A workflow with a function of your own, running on a tool server you start, called from the app itself.
• Stage 5 — Give it a schedule. A nightly pipeline that rolls last night’s incidents up by region — the loop closing back onto the app you started with.
• Stage 6 — Get help. Turn on support for Lumen from the console — no YAML, no redeploy — then answer a user from the triage queue. The first stage about someone other than you.

Two ways to run every command

There are two entry points into Dibbla, and this tutorial serves both:

• The desktop app, where Claude Cowork is preselected in the installer, so a code agent can carry out each step for you.
• The CLI, the dibbla command you install once and drive yourself.

The pages below show the CLI commands at each step regardless of which you use — they are the ground truth the desktop app runs underneath — and each stage also ends with a ready-made prompt you can hand to a code agent.

Start with Stage 1 → Get it live