CF AI Task Assistant

An AI-powered Task Assistant Agent built on Cloudflare's platform. This application demonstrates a complete AI agent system with chat interface, state management, memory, and LLM integration.

Features

• 🤖 AI-Powered Chat: Interactive chat interface using Llama 3.3 (70B) on Workers AI
• 📝 Task Management: Create, update, and manage tasks with status tracking
• 💾 Persistent State: State management using Durable Objects with SQLite
• 🔄 Real-time Communication: WebSocket support for real-time chat
• 🧠 Conversation Memory: Maintains conversation history for context-aware responses
• ⏰ Scheduled Tasks: Automatic cleanup of old tasks

Architecture

This application uses the following Cloudflare services:

1. Workers AI - Runs Llama 3.3 70B model for natural language processing
2. Durable Objects - Provides stateful, persistent storage and WebSocket connections
3. Agents SDK - Framework for building AI agents with built-in state management
4. Workers - Serverless compute for handling HTTP and WebSocket requests

Components

1. LLM Integration

• Model: @cf/meta/llama-3.3-70b-instruct-fp8-fast
• Provider: Cloudflare Workers AI
• Usage: Powers the chat interface and task assistance

2. State & Memory

• Storage: Durable Objects with SQLite backend
• Data: Tasks, conversation history, user preferences
• Persistence: Automatic state synchronization across requests

3. User Input (Chat)

• Protocol: WebSocket for real-time bidirectional communication
• Interface: HTTP endpoint for WebSocket upgrades
• Features: Real-time message streaming and state updates

4. Workflow/Coordination

• Agent Class: Extends Cloudflare Agents SDK
• Scheduled Tasks: Daily cleanup of old completed tasks
• Callable Methods: Exposed API for task management and chat

Project Structure

.
├── src/
│   ├── index.ts          # Worker entry point and routing
│   ├── agent.ts          # TaskAssistantAgent class
│   └── types.ts          # TypeScript type definitions
├── public/
│   └── index.html        # Test HTML page for WebSocket testing
├── wrangler.toml         # Cloudflare Workers configuration
├── package.json          # Dependencies and scripts
├── tsconfig.json         # TypeScript configuration
├── README.md             # This file
└── PROMPTS.md            # AI prompts used in development

Prerequisites

• Node.js 18+ and npm
• Cloudflare account
• Wrangler CLI installed globally (or use npx)

Installation

1. Clone or download this repository

2. Install dependencies:

npm install

3. Authenticate with Cloudflare:

npx wrangler login

Running Locally

Start the development server:

npm run dev

Or using wrangler directly:

npx wrangler dev

The server will start on http://localhost:8787 (or the port shown in the terminal).

Deployment

Deploy to Cloudflare:

npm run deploy

Or:

npx wrangler deploy

After deployment, you'll receive a URL like https://cf-ai-task-assistant.your-subdomain.workers.dev

Usage

WebSocket Connection

Connect to the agent via WebSocket:

const ws = new WebSocket('wss://your-worker-url.workers.dev/?agentId=my-agent');

ws.onopen = () => {
  console.log('Connected to agent');
  
  // Send a chat message
  ws.send(JSON.stringify({
    method: 'chat',
    params: ['Hello! Can you help me manage my tasks?']
  }));
};

ws.onmessage = (event) => {
  const response = JSON.parse(event.data);
  console.log('Agent response:', response);
};

HTTP API

Health Check

curl https://your-worker-url.workers.dev/health

Chat via HTTP (using agent namespace)

curl -X POST https://your-worker-url.workers.dev/api/chat?agentId=my-agent \
  -H "Content-Type: application/json" \
  -d '{"method": "chat", "params": ["Hello!"]}'

Available Methods

The agent exposes the following callable methods:

• chat(message: string) - Chat with the AI assistant
• addTask(title: string, description?: string) - Add a new task
• listTasks(status?: string) - List tasks (optionally filtered by status)
• updateTaskStatus(taskId: string, status: string) - Update task status
• deleteTask(taskId: string) - Delete a task
• getConversationHistory() - Get conversation history
• clearHistory() - Clear conversation history

Example Interaction

1. Connect to the agent via WebSocket
2. Chat: "I need to finish my project report"
3. Agent responds: "I can help you manage that! Would you like me to add 'Finish project report' as a task?"
4. Add task: Call addTask("Finish project report", "Complete the quarterly project report")
5. Check tasks: Call listTasks() to see all tasks
6. Update status: Call updateTaskStatus(taskId, "in_progress") when you start
7. Complete: Call updateTaskStatus(taskId, "completed") when done

Testing Components

Test LLM Integration

# Connect via WebSocket and send a chat message
# The agent will use Llama 3.3 to respond

Test State Management

# Add a task, disconnect, reconnect with same agentId
# The task should still be there (persisted in Durable Object)

Test WebSocket

# Use a WebSocket client to connect and send messages
# Verify real-time bidirectional communication

Test Scheduled Tasks

# Wait for scheduled cleanup (or manually trigger)
# Verify old completed tasks are removed

Configuration

Edit wrangler.toml to configure:

• Worker name
• Durable Object bindings
• AI bindings
• Environment variables

Development Notes

• The agent maintains conversation history for context
• Tasks are stored in the agent's state (Durable Object)
• WebSocket connections are managed by the Agents SDK
• State changes automatically sync to connected clients

Troubleshooting

1. Authentication errors: Run npx wrangler login
2. Build errors: Ensure TypeScript and dependencies are installed
3. WebSocket connection fails: Check that the URL uses wss:// (not ws://)
4. AI model not found: Ensure Workers AI is enabled in your Cloudflare account

License

MIT

Author

Built for Cloudflare AI assignment submission.