Flutter's GenUI Package Overhauled: New Architecture Empowers Developers with Greater Control

Breaking News: Flutter's GenUI and A2UI Protocol Get Major Update

January 2025 - The Flutter team has released a significant update to the genui package and the A2UI protocol, shifting from a 'Structured Output First' philosophy to a 'Prompt First' approach. This change fundamentally alters how developers build generative user interfaces (GenUI) in Flutter.

According to the Flutter team, the update 'decouples the architecture, providing more direct control over how your application interacts with Large Language Models (LLMs).' The removal of the ContentGenerator class is the most impactful code change, forcing developers to take charge of LLM connections and message flows.

'This decoupling means you have control over chat history, retry logic, and error handling,' stated a Flutter engineer involved in the update. 'You’re free to use the model and provider you prefer without going through the framework’s API.'

Background: GenUI and A2UI

Generative UI (GenUI) is a user experience pattern where an agent not only generates content but also decides how to display it and make it interactive. For Flutter developers, implementing GenUI involved using the A2UI protocol—an open standard that defines collaboration between agents and renderers.

The genui package, built by the Flutter team, bridged agents and widgets via A2UI. Earlier versions relied on the ContentGenerator class to hide prompt construction, LLM calls, and response parsing. The latest update eliminates this wrapper, giving developers more freedom.

Architecture Decoupling: From One Class to Three Layers

Previous GenUI versions used a series of classes based on ContentGenerator. The v0.9.0 update removes this entirely. The framework now splits into three distinct layers:

• Engine (SurfaceController) – manages UI state and rendering.
• Transport (A2uiTransportAdapter) – streams messages between agent and renderer.
• Facade (Conversation) – provides a high-level API for chat state management.

This decoupling means provider-specific wrapper packages (like genui_dartantic, genui_google_generative_ai, genui_firebase_ai) are no longer needed. Developers must now set up their own LLM connections and pass messages through a TransportAdapter.

'Instead of passing a ContentGenerator to your SurfaceController, your app is responsible for setting up a connection to an agent and passing messages back and forth through a TransportAdapter,' explained a Flutter documentation contributor.

What This Means for Developers

This architectural shift offers both flexibility and responsibility. Developers can now choose any LLM provider, tweak generation settings, and add custom functions without framework restrictions. However, they must implement their own chat loops, error handling, and retry logic.

The update follows adoption of A2UI protocol v0.9, which switches from streaming structured output APIs to embedding JSON blocks within plain-text agent responses. This 'Prompt First' approach simplifies prompt engineering but requires developers to parse JSON from LLM outputs.

For those migrating from v0.7.0 to v0.9.0, the Flutter team has published a migration guide covering dependency cleanup and chat loop wiring. The key step: remove all references to ContentGenerator and set up a manual TransportAdapter connection.

Migration Guide (Summary)

1. Update genui package to v0.9.0 in pubspec.yaml.
2. Remove deprecated provider packages (e.g., genui_firebase_ai).
3. Replace ContentGenerator with a direct LLM client and TransportAdapter.
4. Implement chat loop using the new Conversation facade.

Full details are available in the official Flutter documentation.

This story will be updated as more developer feedback emerges.