deepset-ai/haystack ? reverse-engineered prompt

Build me a Python framework for creating custom AI applications. The main thing I want is to build systems that can accurately answer questions using a specific set of my own documents, like a production-ready RAG system. It should be able to handle semantic search and question-answering over my data.

Beyond just simple Q&A, I want to create more advanced AI agents that can perform multi-step tasks. It's really important that this is flexible, so I can easily switch between different LLMs from OpenAI, Mistral, Anthropic, or even local models I run myself. I need to have explicit control over the entire workflow—how documents are retrieved, how information is routed and filtered, and how the final answer is generated. The goal is to build powerful, context-aware AI agent workflows, not just a simple chatbot wrapper.