Using A2UI as the application UI layer in Democr.ai

[democr-ai]

Hi everyone,

we wanted to share what we have been building with A2UI in Democr.ai.

Instead of using A2UI only for isolated widgets or small demos, we implemented it as the UI contract for the whole application runtime: modules declare backend-owned UI surfaces, actions, effects, and state transitions, and different clients render those same definitions as native application screens.

Demo video:

a2ui_democrai_sample.mp4

What we added around the protocol:

• backend-declared YAML UI surfaces
• permission-aware properties, so components and actions can be shown, hidden, enabled, or disabled according to runtime policy
• visibility/display metadata for client rendering decisions
• streamed effects from backend actions and agents
• example clients/renderers consuming the same UI contract
• example engines, extractors, and modules showing how the runtime can be extended

The important idea for us is that A2UI is not just a presentation layer. In Democr.ai it becomes part of the application boundary: the backend owns behavior, permissions, data access, actions, and orchestration, while clients focus on rendering and interaction.

The repository includes working examples for:

• clients
• AI engines
• knowledge extractors
• application modules

Those are meant both as examples and as starting points for custom implementations.

Project:

https://github.com/democr-ai/democrai

Website/docs:

https://democr.ai/
https://democr.ai/docs/

We would love feedback from the A2UI community, especially on the protocol extensions around permissions, visibility, effects, and multi-client rendering.