bytedance/deer-flow ? reverse-engineered prompt

I want to build an advanced AI agent system. It shouldn't be just one agent, but a whole framework of them that can work together to handle complex, long-running tasks. For example, I should be able to give it a high-level goal like "research the current state of quantum computing and write a summary report," and the system should be able to figure out the steps, use tools, browse the web, and produce the final document.

It should have a web interface for me to interact with it and a safe sandbox environment where it can write and run code. The system needs to have memory so it can learn from past tasks. Please build this out using the recommended Docker setup. The backend is Python and there's a frontend too. Once it's running, point me to the config file so I can add my API keys for various models like GPT-4 or Gemini.