Design principles

The App Framework is built for mid-maturity developers and data scientists to hit the ground running. These principles guide every decision:

1. Batteries included — not so DRY If a consumer of an application template needs to modify something and that code lives elsewhere, re-evaluate the design. Co-location is preferable to indirection. A sign that this principle is broken is the appearance of pull requests for installable shared libraries.

2. Minimize startup friction Getting started is where users drop off. Make configuration easy, reduce README steps with automation and product features.

3. Rapid iteration is king Fast iteration cycles for the most common customizations make development more effective. Changes to custom models, Flask servers, and Vite applications should be immediate. Set expectations clearly when a workflow cannot be fast.

4. Nerdy is great This framework is built for developers. If it is enjoyable to use, customers likely enjoy it too, unless it becomes too obtuse.

5. DataRobot API token is all you need Everything but AI (and storage, hopefully) is easy to swap out for preferred technology stacks. One credential to rule them all.

6. Make updates easy With many templates and many clones, automated/guided updates are essential. File structure, dependency isolation, and tools like copier are critical for merge-conflict-free updates.

7. Monorepos, not monoliths An application template consists of multiple services, custom models, and notebooks. Each top-level folder must stay completely independent and maintain its own dependency definitions. Use separate dependencies for each area. Modern, fast tooling includes uv, pnpm, and wasm.

8. The project management triangle - App Templates: Fast and Cheap — App authors make them Good. - App Framework: Good and Cheap — App Template authors make them Fast. - CLI: Good and Cheap — delivered fast for App authors.