A2UI-for-Google-Apps-Script

Overview

Bringing A2UI to Google Workspace with Gemini

This repository demonstrates how to implement Google's A2UI (Agent-to-User Interface) protocol entirely within the Google Apps Script (GAS) ecosystem.

By porting the official concepts to GAS, this project enables developers to create dynamic, AI-generated interactive interfaces—such as booking forms, live editors, and metadata viewers—that run natively in Google Sheets Dialogs and Web Apps without complex external server infrastructure.

Description

A2UI is a "secure-by-design" protocol that allows AI agents to generate rich user interfaces using a strict schema rather than arbitrary code. This project leverages the Gemini API to orchestrate these interfaces within Google Workspace, focusing on Task-Driven UIs where the interface adapts to the user's progress.

Key Features:

• Serverless Architecture: Runs entirely on Google Apps Script (GAS).
• Workspace Integration: Interacts directly with Google Sheets, Calendar, and Drive.
• Recursive UI Loop (Sample 4): A stateful workflow where Gemini evaluates previous interactions to generate the next appropriate UI (e.g., Select File -> OK -> Open Editor).
• Feedback-Loop Architecture (Sample 5): Uses Google Sheets as a persistent memory store to track user performance and adaptively adjust content difficulty.
• Gemini Routing: Uses LLM-based routing to select appropriate tools and generate A2UI JSON payloads.

Workflow

The implementation follows a stateful, agent-driven lifecycle. The backend uses the conversation history to decide which functional UI to build next.

1. User Input: The user interacts with an HTML interface.
2. Request: google.script.run securely bridges the client and the GAS backend.
3. Reasoning: Gemini acts as a router to decide which tool to execute and generates an initial UI (e.g., a File Selector).
4. Recursive Interaction: When the user performs an action (like clicking "OK"), the client sends a System Event back to GAS.
5. Stateful Generation: Gemini evaluates the original prompt + new selection from the history to generate the final functional interface (e.g., a dynamic Text Editor).

graph TD
    User((User)) <--> Client[HTML Interface / Sheet Dialog]
    Client <-- google.script.run --> Agent[GAS Server / Gemini API]
    Agent -- 1. Route Intent --> Router{Gemini Router}
    Router -- 2. Execute Tool --> Tools[Apps Script Functions]
    Tools <--> Data[(Sheets / Calendar / Drive)]
    Tools -- 3. Result --> Agent
    Agent -- 4. Generate A2UI JSON --> Client
    Client -- 5. Interaction Event + Selection --> Agent
    Agent -- 6. Re-generate UI based on History --> Client

Loading

Usage

1. Get Gemini API Key

Obtain a valid API key from Google AI Studio.

2. Copy a Sample Project

Select a sample below and copy the Google Spreadsheet to your Drive.

Sample	Type	Description	Link
1. Restaurant Finder	Web App	Official A2UI port. Search & Book restaurants.	Copy
2. Budget Simulator	Web App	Charts & Sheet updates.	Copy
3. Workspace Sync	Dialog	Integrated Sidebar/Dialog with Calendar sync.	Copy
4. Drive Task Agent	Dialog	Recursive Loop! Dynamic file metadata/content editor.	Copy
5. Adaptive Learning	Sidebar	Feedback Loop! Persistent memory & adaptive quizzes.	Copy

3. Setup Script

1. Open the copied Spreadsheet.
2. Go to Extensions > Apps Script.
3. Open main.gs (or Code.gs) and set your API Key to const apiKey = "###";.
4. Save the script.

4. Run (For Dialog/Sidebar Samples 3, 4 & 5)

1. Reload the Spreadsheet.
2. Click the custom menu "sample" > "run" (or "Start Learning").
3. A dialog or sidebar will open. Enter your request.

---

Sample Details

Sample 1: Restaurant Finder (Web App)

This sample reproduces the official A2UI "Restaurant finder" agent as a standalone Web App.

• Action: Enter "Find 3 Chinese restaurants in New York".
• Interaction: The AI returns list cards. Clicking "Book Now" dynamically generates a reservation form.

Initial View	Search Result	Reservation Form

Sample 2: Budget Simulator (Web App)

A practical business automation sample that calculates a household budget and updates a Google Sheet.

• Action: Ask "Check this month's budget".
• Interaction: The AI reads data from the sheet, generates a pie chart (A2UI), and offers a simulation.

Budget Visualization	Simulation Result	Data Updated

Sample 3: Workspace Sync (Dialog)

This sample demonstrates "Bringing A2UI to Google Workspace". It runs inside a modal dialog within Google Sheets and interacts with local data and Google Calendar.

Demonstration Video (YouTube):

Supported Prompts:

1. Restaurant Booking:
   • "Find 3 Chinese restaurants in New York"
   • The agent finds restaurants (mock data) and provides a booking form.
2. Event Management:
   • "Show me events for Jan 17-20"
   • The agent searches the "data" sheet for events, displays them in a checkbox list, and allows you to add selected events to your Google Calendar.

Source Code: sample/A2UI-Google-Sheets

---

Sample 4: Drive Task Agent (Recursive UI Loop)

This sample demonstrates Recursive UI Generation. The agent builds the tool you need based on the conversation history and your interactions.

Demonstration Video (YouTube):

Mermaid Chart Playground

Supported Prompts: Replace 'sample' with the actual folder name in your Google Drive.

1. View Content: "List files in 'sample' and show their content."
2. View Metadata: "Show me metadata for files in 'sample'."
3. Verify and Edit: "I want to edit a file in 'sample'. Show me the files first."

Key Mechanics:

• The "OK" Button: Clicking OK sends a System Event with selected file IDs. Gemini analyzes the original intent to decide whether to render a Viewer, Metadata card, or an Editor.
• Dynamic Editor: Supports real-time text file updates (e.g., .txt, .json, .html).

Source Code: sample/Drive-Task-Agent

---

Sample 5: Adaptive Learning Agent (AI Tutor)

This sample demonstrates a Feedback-Loop Architecture utilizing Google Sheets as a persistent memory store. The agent generates personalized quizzes, tracks performance, and dynamically adjusts difficulty based on historical data.

Demonstration Video (YouTube):

Mermaid Chart Playground

Supported Prompts:

1. Adaptive Learning: "Create a quiz about Google Apps Script basics."
   • Generates targeted questions.
   • Analyzes answers and logs results to Sheets.
   • Provides a "Performance Analysis" dashboard with next steps.
2. Versatility Verification: "Find 3 Chinese restaurants in New York"
   • Demonstrates that the same architecture handles complex transactional flows (booking forms) alongside educational logic.

Key Mechanics:

• Goal Definition & Context Retrieval: Scans the active Spreadsheet for past performance data before generating content.
• Immediate Feedback: Validates answers in real-time and updates the "Memory" (Google Sheets).
• Strategic Analysis: Gemini provides qualitative feedback (Strengths/Weaknesses) rather than just scores.

---

Summary

• Recursive Logic: AI can manage multi-step workflows by generating specialized interfaces on the fly.
• Dynamic UI: A2UI enables AI agents to render interactive components (buttons, charts, forms, editors) based on context.
• GAS-Native: Leverages google.script.run for efficient, serverless client-server communication.
• Workspace Synergy: Seamlessly bridges the gap between AI reasoning and Google Workspace data management.

References

• A2UI: An open project for agent-driven interfaces
• Official A2UI Repository
• Article: A2UI for Google Apps Script
• Article: Bringing A2UI to Google Workspace with Gemini
• Article: Beyond Chatbots: Building Task-Driven Agentic Interfaces in Google Workspace with A2UI and Gemini

---

Licence

MIT

Author

Tanaike

TOP

---

Update History

• v1.0.0 (January 8, 2026)

  • Initial release.

• v1.0.1 (January 19, 2026)

  • Added a sample 3 "Workspace Sync".

• v1.0.2 (February 3, 2026)

  • Added a sample 4 "Drive Task Agent" with recursive UI generation logic.

• v1.0.3 (February 7, 2026)

  • Added a sample 5 "Adaptive Learning Agent" with feedback-loop architecture.