Agent UI Starter Pack (A2UI)

High-Fidelity Agent-Driven User Interfaces for Google Cloud.

The Agent UI Starter Pack is a professional distribution for developers building high-fidelity AI applications on Gemini. We provide the architectural "Golden Path" for bridging the gap between conversational intelligence and actionable software.

💡 The Core Mission

❌ The Problem: The "Wall of Text"

Conversational AI today is highly intelligent but low-utility. Users are often stuck behind a "Wall of Text"—trying to parse complex stats, project roadmaps, or financial data out of a chat bubble. This leads to Text Fatigue and limits the agent's ability to act as a real tool.

✅ The Solution: From Chatbot to Cockpit

We move users away from talking to a "box" and into an Agent Cockpit. Instead of just sending text, your agent "manifests" high-fidelity, interactive UI components on the fly using the A2UI Protocol.

Software isn't static anymore; it's synthesized by the Agent based on the user's intent.

🎁 What do you get?

By using this starter pack, you aren't just getting template code—you are getting a Production Framework:

• Instant 0-to-1: Skip 3 weeks of setting up Vertex AI Auth, NL-to-JSON parsing, and dynamic React rendering.
• The Artifact Registry: A library of 20+ premium React components (StatBars, QuizCards, Timeline) that are native to JSON.
• The Bridge Orchestrator: A specialized backend that triages "Conversational Intent" and resolves which UI surfaces to manifest.
• Observability-by-Default: A built-in "Ops Console" to inspect the raw NDJSON thought-process of the agent in real-time.

🏗️ Core Pillars

🎭 The Face (Front End) - Primary Focus

Role: The Experience. Adaptive surfaces that change based on what the agent is doing.

• CLI: agent-ui-starter-pack
• Powered by: React, Vite, A2UI Protocol.
• Feature: Dynamic A2UI Renderer and a library of high-fidelity components.

⚙️ The Engine (Agent)

Role: The Brain. Internal reasoning and tool execution.

• CLI: agent-starter-pack
• Powered by: Python, Vertex AI SDK, ADK.
• Feature: Native integration with Agent Engine for managed runtimes.

🚀 Key Features

💎 A2UI Protocol Native

The entire stack is built on the Agent-Driven User Interface (A2UI) protocol. Your agent doesn't just send text; it sends structured JSON that manifests as premium UI components (Timelines, Trophies, Quizzes) in real-time.

🔄 State Synchronization

Seamlessly sync agent reasoning steps with frontend state. Build "Human-in-the-loop" workflows where the user can inspect and refine agent actions before they finalize.

🛠️ Usage (Prescribed Examples)

Scaffolding

Create a new project in seconds using our specialized CLIs:

To create the High-Fidelity Front End:

uvx agent-ui-starter-pack create my-ui-project

To hydrate a Figma wireframe into A2UI:

uvx agent-ui-starter-pack hydrate <figma-url>

Local Development

Start the integrated Vite + API bridge:

make dev

Production Deployment

Deploy the full stack to Google Cloud:

make deploy-prod

📁 Repository Structure

• /src/a2ui: Core A2UI rendering logic and base components.
• /src/backend: The agent engine logic and API bridge.
• /src/components: Premium UI building blocks.
• /docs: Detailed integration guides and protocol specifications.

License

MIT