oobabooga/textgen ? reverse-engineered prompt

Build me a desktop app for running local AI models on my own machine, with a clean chat window and a simple way to load models from a local folder. It should work fully offline, with no telemetry or external calls unless I turn on things like web search. I want chat and instruction modes, support for attaching images and documents like text files, PDFs, and docx files, and the ability to edit messages and branch conversations. Include a notebook style area for freeform generation too.

Please also make it expose a local API that is compatible with OpenAI and Anthropic style requests, including tool calling so models can use custom functions like web search, page fetching, and math. If possible, support switching between model backends without restarting, and make the app easy to launch on Windows, macOS, and Linux. If you need to check current docs for any implementation details, go ahead and look them up online.