ItzCrazyKns/Vane ? reverse-engineered prompt

I want to build a self-hosted, privacy-focused AI answering engine, similar to Perplexity but something I can run myself. It should have a clean, modern interface where I can ask a question and get a direct, formatted answer with citations from its web sources, not just a list of links.

The most important thing is flexibility with the AI models. It should let me connect to local LLMs running on my machine through Ollama, but also allow me to plug in API keys for cloud providers like OpenAI, Groq, or Anthropic's Claude. For the underlying web search, please use SearxNG to keep everything private.

It should have different search modes—like a quick mode and a more thorough one for research. Users should also be able to upload their own documents, like PDFs, and ask questions about them. Oh, and add some helpful widgets for things like weather or stock prices. Make it all easy to deploy using Docker and build it with a modern stack like Next.js and TypeScript.