cheumk/aimodel

cf_ai_intelligent-assistant

AI chat assistant built on Cloudflare's platform using Llama 3.3, with conversation memory powered by Durable Objects.

Live Demo

• Chat UI: https://cf-ai-assistant-76b.pages.dev

What it does

This is a chat application where you can have conversations with Llama 3.3. The conversations are saved using Cloudflare Durable Objects, so when you refresh the page your chat history is still there. Everything runs on Cloudflare's edge network.

Tech Stack

• Workers AI - runs Llama 3.3 70B model
• Cloudflare Workers - API backend
• Durable Objects - stores conversation history (up to 50 messages per session)
• Cloudflare Pages - hosts the chat UI

Running Locally

Install dependencies:

npm install

Start the Worker API (runs on port 8787):

npm run dev

The Pages dev command has issues with Durable Objects locally, so just serve the HTML with Python:

cd public
python -m http.server 8788

Then open http://localhost:8788 in your browser.

Deploying

Auto-Deploy with GitHub Actions (Recommended)

The project auto-deploys on every push to main branch.

Setup:

1. Get your Cloudflare API Token:

   • Go to https://dash.cloudflare.com/profile/api-tokens
   • Create a token with "Edit Cloudflare Workers" permissions

2. Add the token to GitHub:

   • Go to your repo Settings → Secrets and variables → Actions
   • Create a new secret named CLOUDFLARE_API_TOKEN
   • Paste your API token

3. On first deployment, create the Pages project manually:

   npx wrangler login
   npx wrangler pages project create cf-ai-assistant --production-branch=main
   

4. Commit and push to main - it will auto-deploy!

Manual Deploy

First login to Cloudflare:

npx wrangler login

Deploy the Worker:

npm run deploy

Create and deploy the Pages project:

npx wrangler pages project create cf-ai-assistant --production-branch=main
npx wrangler pages deploy public --project-name=cf-ai-assistant

After deploying, update the Worker URL in public/index.html (line 349) to match your deployed Worker URL, then redeploy Pages.

API Endpoints

POST /api/chat

Send a message to the AI.

{
  "message": "What is Cloudflare?",
  "conversationId": "default",
  "userId": "user123"
}

Returns the AI's response along with conversation ID and timestamp.

GET /api/history

Get conversation history.

Query params: conversationId (default: "default"), limit (default: 20)

POST /api/clear

Clear conversation history.

{
  "conversationId": "default"
}

GET /api/health

Health check - returns status and timestamp.

Project Structure

src/
  index.ts              - Worker API with AI integration
  conversation-state.ts - Durable Object for state management
public/
  index.html           - Chat interface
wrangler.toml          - Cloudflare configuration

How it Works

1. User sends a message through the chat UI
2. Worker API receives the message and fetches conversation context from the Durable Object
3. Context + new message is sent to Llama 3.3 via Workers AI
4. AI response is saved to Durable Object along with the user's message
5. Response is returned to the UI

The Durable Object keeps the last 10 messages as context for the AI, so it can have coherent multi-turn conversations.

Customization

Change the AI model - Edit src/index.ts line 69. See available models at https://developers.cloudflare.com/workers-ai/models/

Adjust system prompt - Edit src/index.ts line 66

Change message limit - Edit src/conversation-state.ts line 72 (currently 50 messages)

Modify UI styling - All styles are in public/index.html

Notes

• Workers AI charges apply even in local development (it connects to Cloudflare's API)
• Free tier includes 10,000 AI requests/day
• The Durable Object migration uses new_sqlite_classes for free tier compatibility
• Conversation history persists across deployments

Troubleshooting

"AI binding not found" - Check wrangler.toml has the [ai] binding

Durable Objects error on deploy - Make sure wrangler.toml uses new_sqlite_classes in the migration (not new_classes)

CORS errors - The Worker has Access-Control-Allow-Origin: * set for all responses

Pages can't find Worker - Update the API_BASE URL in public/index.html to match your deployed Worker

License

MIT - see LICENSE file

---

Built for Cloudflare AI assignment. See PROMPTS.md for details on AI-assisted development.