Ollama's 0.33 preview lets you switch local models on and off inside Claude's desktop app

The 0.33 release candidate, published 21 August, adds a menu-bar control for choosing which local models Claude may use, and an Apps view for managing integrations. It also repairs cache behaviour: a cancelled prefill now keeps every restore point it crossed, so a retry resumes where it stopped, and Ollama switched off Claude Code’s tokens-left message, which it had been moving to the front of the prompt and thereby invalidating the cache on every single request.

What it means One misplaced system message was busting the cache on every request — worth checking how your own prompt is assembled before blaming the model for being slow.