TL;DR: Ollama released v0.33.0, adding per-model controls in Claude Desktop and fixing KV-cache-breaking prompt handling for AI builders.
Summary: Ollama released v0.33.0, introducing a menu-bar toggle to turn individual local models on or off for Claude Desktop and letting Claude pick from available Ollama models. The update also overhauls prefill caching so cancelled requests resume from trustworthy restore points, and it disables Claude Code's token-countdown system message that was invalidating the KV cache on every request.
Why it matters: For AI builders using Claude Code with local Ollama models, this removes a KV-cache-breaking prompt hack and makes retries resume where they stopped, cutting wasted compute and latency. Update Ollama and test the new prefill restore behavior with long-context requests.
Source: github_releases