h2oai/h2ogpt ? reverse-engineered prompt

Build me a private local GPT app I can run on my own machine or in Docker, where I can chat with the model and ask questions about my files. I want to upload documents like PDFs, Word files, spreadsheets, images, audio, video, and text, then have it search, summarize, and answer questions from that content in a private way.

Please make it support a simple web interface where I can chat, upload files, see my collections, and stream responses as they come in. It should work with local models too, not just hosted ones, and it would be great if it could talk to common model backends and expose an OpenAI compatible API so other tools can use it. If it makes sense, add voice input and output, basic image understanding, and a way to run it on Linux, macOS, Windows, or Docker. Keep it polished enough that I can start it up and use it without a lot of setup.