genai_primitives

Generative UI SDK for Flutter (genui)

A Flutter library to enable developers to easily add interactive generative UI to their applications.

See the Getting started with GenUI video for an overview of the package.

Status: highly experimental

This is a highly experimental package, which means the API will change (sometimes drastically). Feedback is very welcome.

Summary

Our goal for the GenUI SDK for Flutter is to help you replace static "walls of text" from your LLM with dynamic, interactive, graphical UI. It uses a JSON-based format to compose UIs from your existing widget catalog, turning conversations or agent interactions into rich, intuitive experiences. State changes in the UI update a client-side data model, which is fed back to the agent, creating a powerful, high-bandwidth interaction loop. The GenUI SDK for Flutter aims to be easy to integrate into your Flutter application to significantly improve the usability and satisfaction of your chatbots and next-generation agent-based user experiences.

High-level goals

• Increase Interaction Bandwidth for Users: allow users to interact with data and controls directly, making task completion faster and more intuitive. Move beyond a "wall of text".
• Simple and easy to use by developers: seamlessly integrate with your existing Flutter workflow, design systems, and widget catalogs.
• Drive agent-human UX forward: innovate ways to dramatically improve how users interact with their LLMs and Agents. Radically simplify the process of building UI-based agent interactions, by eliminating custom middleware between the agent and the UI layer.

Features

• Multiple agent and LLM provider support
• A2UI support
• Standard UI catalog
• Custom widgets
• Data binding
• Chat UX
• Canvas UX

Use cases

• Incorporate graphical UI into chatbots: instead of describing a list of products in text, the LLM can render an interactive carousel of product widgets. Instead of asking for a user to type out answers to questions, the LLM can render sliders, checkboxes, and more.
• Create dynamically composed UIs: an agent can generate a complete form with sliders, date pickers, and text fields on the fly based on a user's request to "book a flight."

Look & feel

The GIF above shows how GenUI enables dynamic, interactive UI generation, instead of text descriptions or code from a traditional AI coding agent.

Core difference

This UI is not generated in the form of code; rather, it's generated at runtime based on a widget catalog from the developers' project.

Implementation goals

• Integrate with your LLM: Work with your chosen LLM and backend to incorporate graphical UI responses alongside traditional text.
• Leverage Your Widget Catalog: Render UI using your existing, beautifully crafted widgets for brand and design consistency.
• Interactive State Feedback: Widget state changes are sent back to the LLM, enabling a true interactive loop where the UI influences the agent's next steps.
• Framework Agnostic: Be integrated into your agent library or LLM framework of choice.
• JSON Based: Use a simple, open standard for UI definition—no proprietary formats.
• Cross-Platform Flutter: Work anywhere Flutter works (mobile, iOS, Android, Web, and more).
• Widget Composition: Support nested layouts and composition of widgets for complex UIs.
• Basic Layout: LLM-driven basic layout generation.
• Any Model: Integrate with any LLM that can generate structured JSON output.

Connecting to an AI agent

The genui framework is designed to be backend agnostic. You can use any AI SDK (such as firebase_ai or dartantic_ai) to generate content. The framework provides adapters (like A2uiTransportAdapter) to ingest the AI response and render it.

For custom agent servers that implement the A2UI protocol, you can use the genui_a2a package.

See the package table below for more details on each.

Packages

Package	Description	Version
genui	The core framework to employ Generative UI.
genui_a2a	Provides A2uiAgentConnector for connecting to any server that implements the A2UI protocol. Use this for integrating with custom agent backends.
genai_primitives	A set of technology-agnostic primitive types and data structures for building Generative AI applications.
json_schema_builder	A fully featured Dart JSON Schema package with validation, used by the core framework to define widget data structures.

Dependencies

This diagram shows how packages depend on each other and how examples use them.

graph TD
  dev_tools/catalog_gallery --> genui
  examples/simple_chat --> genui
  dev_tools/composer --> genui
  examples/verdure --> genui_a2a
  genui_a2a --> genui
  genui --> genai_primitives
  genai_primitives --> json_schema_builder

A2UI Support

The Flutter Gen UI SDK uses the A2UI protocol to represent UI content internally. The genui_a2a package allows it to act as a renderer for UIs generated by an A2UI backend agent, similar to the other A2UI renderers which are maintained within the A2UI repository.

The Flutter Gen UI SDK currently supports A2UI v0.9.

Getting started

See the genui getting started guide.

Skills

This repo contains skill files for developers building with agentic coding tools. They can be copied directly into an agent's preferred location or installed using the skills package:

npx skills add https://github.com/flutter/genui/tree/main/packages/genui/skills

Constraints

This repo requires Flutter version >=3.35.7.

Some things we're thinking about

• Genkit Integration: Integration with Genkit.
• ADK Plugin: turnkey integration with ADK.
• Expanded LLM Framework Support: Official support for additional LLM frameworks.
• Streaming UI: Support for progressively rendering UI components as they stream from the LLM.
• Full-Screen Composition: Enable LLM-driven composition and navigation of entire app screens.
• A2A Agent Support: Support for A2A agent interactions.
• Dart Bytecode: Future support for Dart Bytecode for even greater dynamism and flexibility.

Contribute

See CONTRIBUTING.md