You link your os to a local or cloud llm, and a local program asking the OS for a response and can’t even tell which one you’re using or whether it’s on the machine or not. It should all be abstracted away.
There's a number of standard APIs already, OpenAI supports Anthopic's MCP, LM studio supports both their proprietary API as well as OpenAI's API. OpenAI has open sourced their realtime API (https://github.com/openai/openai-realtime-console/tree/webso...) and others. Most local clients just have a https://URL:port and then a drop down box for which RESTful API you want to use (for 88% of use cases they all support the same stuff, for realtime it's not quite settled yet), plus a field for an API key if needed.
To me, the value of these types of projects is specifically that they are self-contained and local-only. That's the only kind of interaction with it I'm comfortable with right now. I mostly jumped ship on commercial software a long time ago, so I'm hoping there will still be some AI-free linux distros for a good long time. Different strokes for different folks, I suppose. At the point that the type of AI integration you're imagining becomes ubiquitous and mandatory, I may or may not stop working with computers entirely, depending on the state of the tech and the state of society by then.
You link your os to a local or cloud llm, and a local program asking the OS for a response and can’t even tell which one you’re using or whether it’s on the machine or not. It should all be abstracted away.