Cloudflare RAG

Minimal RAG application on Cloudflare: Worker (TypeScript) + Workers AI (LLM) + D1 (documents) + KV (conversation memory) + React frontend.

Assignment mapping

• LLM: Workers AI (@cf/meta/llama-3.1-8b-instruct)
• Workflow/coordination: Cloudflare Worker (worker/src/rag-worker.ts)
• User input: React chat UI (frontend/src)
• Memory/state: KV (CONVERSATIONS) for conversation history

Live demo

• API (Worker): https://rag-worker.tchan-efa.workers.dev
• Frontend (Pages): https://819dfebf.rag-llm-aqj.pages.dev

Data

This app returns sources when matching content exists in D1. If you fork this repo, load your documents into D1 using the scripts in migrations/.

Minimal commands

# Insert a single test document (production D1)
wrangler d1 execute DB --command "INSERT INTO documents (id, content, source, metadata) VALUES ('test1','Reinforcement learning is a type of machine learning where an agent learns by interacting with an environment and receiving rewards.','test.md','{\"page\":1}');"

# Verify document count
wrangler d1 execute DB --command "SELECT COUNT(*) AS total FROM documents;"

Structure

cloudflare-rag/
├── worker/            # Cloudflare Worker API
│   ├── src/           # RAG pipeline
│   └── wrangler.toml  # Bindings (AI, D1, KV, Vectorize)
├── frontend/          # React chat UI
└── migrations/        # Import scripts

Repository requirements

• Repository name must start with cf_ai_
• Include PROMPTS.md at repo root with prompts used
• Include this README with clear run instructions