LyalinDotCom/GemmaDesktop — reverse-engineered prompt

Build me an experimental local Gemma workspace app for Mac. I want a polished desktop app where I can chat with a local model through Ollama, point it at a project folder, browse files, ask it to help with coding, research web pages, read PDFs and attachments, handle images or audio when possible, and show clear diagnostics when the model or runtime is not working.

Also make a terminal version for coding tasks, so I can ask it to inspect a repo, edit files, run commands, remember a session, and validate the build. Keep shared model, tool, session, attachment, and runtime logic in one reusable SDK so the desktop app and terminal app behave consistently.

Please make it local by default, honest about alpha risks, and easy to run with npm. Include checks or tests for the important behavior. Use Gemma through Ollama as the main path, but leave room for LM Studio, llama.cpp style servers, and OpenAI compatible local runtimes. Look up current docs online if you need to.