A specialized, event-driven multi-agent Google Workspace Add-on built for the Google Rapid Agent Hackathon. This suite leverages the Google Agent Development Kit (ADK) and Model Context Protocol (MCP) to provide intelligent email management, deep analytics, and self-observing LLM performance tuning.

The system is built on a serverless architecture using Google Cloud Run, orchestrated via Pub/Sub and EventArc triggers. It features three primary agents working in concert:

• Responsibility: Real-time email categorization, automated drafts, and semantic bucket grouping.
• Tech Stack: Voyage AI (embeddings), MongoDB (operational vector cluster/change streams).
• Workflow: Event-driven via Google Cloud Pub/Sub.

• Responsibility: Analytical RAG, computing email velocity patterns, and response metrics.
• Tech Stack: Fivetran, BigQuery, Vertex AI Vector Search.
• Workflow: SQL interaction via BigQuery and semantic parsing via Vertex AI.

• Responsibility: LLM observability, trace diagnostics, and automated prompt performance tuning.
• Tech Stack: OpenInference Tracing SDK, Arize Phoenix, Gemini 3 (LLM-as-a-Judge).
• Workflow: Evaluates trace logs and updates MongoDB semantic notes.

• agents/: Contains the logic for the three specialized agents.
• publish/sidebar-ui/: Gmail Sidebar UI components (Apps Script/TypeScript).
• shared/: Single source of truth for MCP schemas and common utility clients.

• Language: Python 3.11+ (Agents), TypeScript (Sidebar UI)
• Frameworks: Google Agent Development Kit (ADK), Model Context Protocol (MCP)
• Infrastructure: Google Cloud Run, Pub/Sub, EventArc
• Databases: MongoDB, BigQuery
• AI/ML: Gemini 3, Voyage AI, Vertex AI

The Sidebar UI is organized into several key functional screens:

• Main Card: Entry point showing all agent modules.
• Smart Email Manager: Controls for label thresholds and semantic re-organization.
• Inbox Analytics: Detailed metrics, response time analysis, and natural language mailbox querying.
• Self Arize-ing Manager: Central hub for user personalization, data privacy (Delete Data), and MCP connection management.
• User Personalization: Interactive dashboard to review and toggle identified email routines and agent-learned preferences.

The Google Apps Script environment is stateless. The Sidebar UI manages state (e.g., active preference IDs, search queries, or chat history) by serializing data into action parameters using CardService.newAction().setParameters(). This ensures a seamless interactive experience across card updates.

Deployment is handled via serverless containers on Google Cloud Run. See publish/deploy.sh for deployment scripts.

To get the Gmail Sidebar UI running, follow these steps:

Go to script.google.com/home/settings and toggle Google Apps Script API to ON.

Run these commands in the publish/ directory to link your local code to a Google Apps Script project:

cd publish
npm install
npx clasp login
npx clasp create --title "Rapid Agent Gmail Suite" --type gmail --rootDir .

Once linked, use the provided scripts to push code and generate a Deployment ID for the GCP Console:

On Windows:

powershell.exe -File publish/deploy.ps1

On Mac/Linux:

bash publish/deploy.sh

• Rahul Gautham Putcha: Lead Agent Architect & Repo Holder. Responsible for the core agent logic in smart-email-manager and inbox-analytics, MCP schema design, and backend infrastructure. Responsible for the Gmail Sidebar UI in publish/sidebar-ui/.
• Sree Harshitha Jonnalagadda: Data Integration & Observability Specialist. The self-arizeing-manager telemetry pipeline, and project orchestration.

This project is licensed under the MIT License. See the LICENSE file for details.

Notice: This repository has been initialized with a README and LICENSE file. Please update the copyright placeholders in the LICENSE file with your information.