Dart package
Seisei
A little more certainty for AI.
ABOUT THE PROJECT
A little more about Seisei.
Seisei explores a small but useful idea: model output should be something an app can check before using it. It is an early set of Dart and Flutter packages for typed generation, validation, and testing.
01 / Seisei
Say what the app expects.
A schema describes the response shape. In the offline example, a summary must contain a text field. The response passes through validation before the app turns it into its own Dart type.
- Model response
- Schema validation
- App data
const schema = ObjectSchema(
name: 'Summary',
requiredStringFields: {'text'},
);Excerpt from the offline CLI example · Validation can reject a response.
02 / Seisei
Test without calling a model.
A fake provider makes it possible to exercise the same contracts with a known response. The basic CLI example runs offline, so it can check the app’s handling without depending on a live model service.
- Fake provider
- Known response
- Repeatable test
Offline example · No Apple or cloud model call.
Where things stand.
This is experimental software. The core stays provider-neutral, with Apple integration and other adapters in separate packages. It is not a complete AI backend.
Explore Seisei ↗Read the source ↗