An intelligent, self-improving research system powered by Cloudflare Workers and OpenAI. This system uses two Durable Objects to recursively generate, verify, and refine AI research responses.

• Iterative Research: Uses multiple AI passes to refine responses
• Fact Checking: Verifies information through a dedicated fact-checking phase
• Edge Computing: Built on Cloudflare Workers for global performance
• Durable Objects: Maintains consistency across research iterations

• Node.js or Bun (latest version)
• Cloudflare Workers account
• OpenAI API key

1. Clone and install dependencies:

   git clone https://github.com/acoyfellow/ai-research-agent.git
   cd ai-research-agent
   npm install # or bun install

2. Copy wrangler.toml.example to wrangler.toml and configure:

   name = "ai-research-Agent"
   main = "src/index.ts"
   compatibility_date = "2024-02-01"
   
   [vars]
   OPENAI_API_KEY = "your-key-here"
   OPENAI_MODEL = "gpt-4o-mini"
   MAX_ITERATIONS = "5"
   CONFIDENCE_THRESHOLD = "0.9"

3. Deploy:

   npm run deploy # or bun run deploy

Send a POST request to start research:

curl -X POST https://<your-worker>.workers.dev/start \
  -H "Content-Type: application/json" \
  -d '{"topic": "Explain quantum computing in simple terms"}'

{
  "research": "Quantum computing explained in simple terms...",
  "iteration": 2
}

1. Initial Research: ResearchFetcherDO generates the first research pass using OpenAI
2. Fact Checking: FactCheckerDO verifies and improves the research
3. Iteration: Process repeats until max iterations or confidence threshold is met
4. Response: Returns the final refined research

1. Start local development:

   npm run dev # or bun run dev

2. Visit http://localhost:8787 to use the web interface

MIT License - see LICENSE for details

• Star this repo
• Report issues
• Submit PRs
• Share with others

Built with ❤️ using Cloudflare Workers and OpenAI

Inspired by sunil pai @threepointone