A conversation inside the desktop app
Home can use a model already exposed by Ollama, LM Studio or llama.cpp on a local loopback endpoint. The app lists reachable models rather than silently selecting a cloud fallback.
Context stays tied to the selected project
Ask about the approved project and include a current file, selected code or an existing diagnostic when relevant. Bounded context and returned sources help you understand what informed the response.
Edit proposals are an explicit choice
Allow edits opts the next message into the safe project-edit path and resets after sending. Inspect the resulting History state and verify the file outcome rather than assuming model text changed the project.
Set up the right AI path
Start the local runtime and choose an available model for Home. Configure Connect separately when you want an external MCP client such as ChatGPT to work with the project.