Thanks for the heads up. I had a brainstorming session with Gemini about this (I can’t believe I just typed that sentence lol) and the plan is to switch to two local LLMs; a lightweight, fast 3B model for autocomplete, and a slower 14B model for chat sessions. Then I can switch to a DeepSeek Premium API key for the really tough stuff. It recommended the Continue plugin for VSCode.

When I am away from home I’ll run autossh on my dinosaur road laptop (which probably has 8MB video RAM lol) to connect to the home PC’s LLMs. Gemini assured me that this should run well over my intermittent cellular connection.

You just saved me some headache and money :-D