CodePilot — AI Code Optimization Assistant

A Cloudflare-native AI chat application for explaining and improving the performance of Python, C++, and CUDA code. Not a benchmarking or code-execution service: suggestions are hypotheses until measured.

Live Demo

Try CodePilot here: https://ai-code-optimization-assistant.codepilot-ai.workers.dev

Features

• User-submitted code or questions through a responsive chat interface
• Llama 3.3 70B on Cloudflare Workers AI
• Cloudflare Worker coordinates model calls, conversation lookup, input validation, and persistence
• D1 SQL database stores conversation history across refreshes, scoped to a browser session cookie
• Follow-up context (up to 12 previous messages) and reset conversation action
• Plain-text model display prevents generated HTML from executing

Cloudflare assignment requirements

Requirement	Implementation
LLM	Workers AI / Llama 3.3
Coordination	Worker fetch handler
User input	Responsive web chat, served as static Worker assets
Memory/state	D1 persisted messages and session cookie

Local setup

Requires Node.js 20+, npm, and a Cloudflare account.

1. npm install
2. npx wrangler login
3. npx wrangler d1 create code-coach-db
4. Copy the new database_id into the D1 entry in wrangler.jsonc (replace REPLACE_WITH_YOUR_D1_DATABASE_ID). Never deploy with the placeholder.
5. npm run db:local
6. npm run dev
7. Visit the local URL shown by Wrangler.

Deploy

1. npm run db:remote (creates the tables in the production D1 database)
2. npm run deploy
3. Open the deployed *.workers.dev URL printed by Wrangler.

Important: Workers AI usage may incur charges beyond free allowances. Review your account's Cloudflare usage and pricing. Production deployments should add rate limiting and user authentication; this demo deliberately uses only an anonymous, per-browser session cookie. The cookie is not a replacement for sign-in and should not store sensitive code. The app does not execute submitted code.

Testing

• npm run check checks JavaScript syntax.
• npm test runs lightweight API tests.
• After binding the services, test chat, refresh history, and reset in the browser.

Architecture

Browser chat (HTML/CSS/JS)
       | fetch('/api/chat')
Cloudflare Worker
       |-- D1 -> read last 12 messages, store exchange
       `-- Workers AI -> Llama 3.3 70B

Prompt history

See PROMPT_HISTORY.md. The user must add any later AI-assisted edits and prompts they use before submitting. The record documents the request that led to this initial generation, rather than fabricating prompts.

Documentation

• Cloudflare Workers AI model
• D1 setup
• Worker static assets