🧠 Multi-Agent Productivity Orchestration System

A production-grade, multi-agent AI system that autonomously manages tasks, schedules, and knowledge.

---

🌟 Why This Project is Unique

Most AI tools are reactive—you ask a question, and they answer. This system is proactive and orchestrational.

When you provide a high-level goal (e.g., "Prepare for my machine learning exams"), the Orchestrator Agent breaks it down into a dependency graph of subtasks and delegates them to specialized agents:

• 📝 Task Agent: Executes specific actionable items and summarizes outcomes.
• 📅 Calendar Agent: Estimates durations and schedules blocks of time.
• 📚 Knowledge Agent: Extracts, synthesizes, and stores key facts into a vector-ready knowledge base.

Note: While the original design requested Python (FastAPI/SQLAlchemy/Celery), this implementation has been built using TypeScript, Express, and React to ensure it runs flawlessly and interactively within the Google AI Studio preview environment. The architectural patterns (Orchestrator, Tool Abstraction, Async Execution) remain identical.

---

🏗️ Architecture

User Input ("Prepare for exams")
       │
       ▼
[ Orchestrator Agent ] ──► Breaks down goal into JSON Plan
       │
       ├─► [ Task Agent ]      ──► Executes tasks & writes to DB
       ├─► [ Calendar Agent ]  ──► Schedules tasks & writes to DB
       └─► [ Knowledge Agent ] ──► Extracts facts & writes to DB
       │
       ▼
[ Database Layer ] (Tasks, Schedules, Notes, Agent Logs)
       │
       ▼
[ React Dashboard ] (Real-time monitoring of agent activity)

---

✨ Features

• Multi-Agent Coordination: Agents communicate via structured JSON and share a common state.
• Async Workflow Engine: The Orchestrator delegates tasks asynchronously, simulating a background worker queue.
• Real-Time Dashboard: Monitor agent thought processes, task statuses, and generated knowledge in real-time.
• Production-Ready Backend: Built with Express.js, featuring clean API endpoints and robust error handling.
• Cloud-Native: Fully Dockerized and ready for Google Cloud Run deployment.

---

🚀 API Endpoints

POST /api/plan

Accepts a high-level goal and kicks off the multi-agent orchestration.

curl -X POST http://localhost:3000/api/plan \
  -H "Content-Type: application/json" \
  -d '{"goal": "Research quantum computing basics"}'

GET /api/status

Returns the current state of the system (Tasks, Schedules, Notes, Logs).

---

☁️ Google Cloud Deployment

This project is fully containerized and ready to be deployed to Google Cloud Run.

1. Build the Docker Image

export PROJECT_ID="your-gcp-project-id"
export REGION="us-central1"
export IMAGE_NAME="gcr.io/$PROJECT_ID/multi-agent-system"

docker build -t $IMAGE_NAME .

2. Push to Google Container Registry (or Artifact Registry)

gcloud auth configure-docker
docker push $IMAGE_NAME

3. Deploy to Cloud Run

gcloud run deploy multi-agent-system \
  --image $IMAGE_NAME \
  --platform managed \
  --region $REGION \
  --allow-unauthenticated \
  --set-env-vars="GEMINI_API_KEY=your_api_key_here,NODE_ENV=production"

---

📂 Folder Structure

.
├── Dockerfile                  # Production Docker configuration
├── package.json                # Dependencies and scripts
├── server.ts                   # Express Backend Entry Point
├── src/
│   ├── agents/                 # AI Agents
│   │   ├── orchestrator.ts     # Planner & Controller
│   │   ├── task_agent.ts       # Task Execution
│   │   ├── calendar_agent.ts   # Scheduling
│   │   └── knowledge_agent.ts  # Information Extraction
│   ├── api/
│   │   └── router.ts           # FastAPI-equivalent routing
│   ├── db/
│   │   └── schema.ts           # Database schema & ORM abstraction
│   ├── App.tsx                 # React Dashboard
│   └── main.tsx                # Frontend Entry Point

---

🚀 Future Scalability Notes

• Database: Swap the in-memory schema.ts with Prisma + PostgreSQL for persistent storage.
• Message Queue: Integrate BullMQ or RabbitMQ for robust retry logic and distributed agent execution.
• Vector Search: Connect the Knowledge Agent to Pinecone or Milvus for semantic retrieval of stored notes.
• Plug-and-Play Agents: The assignedAgent routing in the Orchestrator can easily be expanded to support new specialized agents (e.g., EmailAgent, CodeAgent).