Build an application with AI

Goal: turn a reviewed model and a written description into an application using the experimental Spec-Driven Agent.

Before you start

Open the editor and create a project. AI availability depends on the deployment: a configured free tier, a sponsored tier, or your own provider key can power a run. A deterministic generator remains available when you do not need AI customisation.

Describe, review, generate

  1. Ask the assistant to model a small system, for example: “Create a book catalogue with titles, authors, and a screen for adding and listing books.”

  2. Review the class and GUI diagrams. Correct names, relationships, and fields before requesting code.

  3. Ask: “Generate the web app.” Creating the model does not itself start a generation run.

  4. Choose a provider if prompted. A saved provider key can start the requested run immediately and incur provider charges; set the per-run budget in the key dialog before starting.

  5. Follow the run card. Use Stop to cancel or wait for the result.

  6. Download the archive and read its README. Review its verification findings before running or deploying it.

What the run does

The agent starts from a BESSER scaffold when one fits the request. For another stack it plans the project structure. It then edits files, validates the result, and attempts repairs within the run’s turn, cost, and time limits.

The checks cover different things: model validation checks the diagram; generated-code validation checks the files; authorised runtime probes check whether a supported backend boots and creates records. Disabled or unavailable checks are recorded as unverified. Read the result’s findings even when the editor presents the application as ready.

Continue your work

Ask for a change in the same project to modify the previous application’s files while they remain available on the server. For longer-term work, push the result to GitHub and later choose Continue from GitHub in the editor.