Ollama’s latest release candidate, v0.33.0-rc2, brings a handful of meaningful updates for self-hosted users, including tighter Claude Desktop integration and several under-the-hood improvements that make model caching more dependable.
The Claude Desktop app now works directly with Ollama. You can toggle individual Ollama models on or off right from the menu bar, and pick any of your available local models from within Claude without needing to fiddle with configuration files. Cloud models only appear when you’re signed in, keeping the local and remote options clearly separated. There’s also a new Apps view that manages app integrations and provides copyable commands, making setup more straightforward.
Caching fixes are a big part of this release. Previously, agent clients that cancelled long text generations could hang. Now, restore points for prefill operations are trustworthy by design: a cancelled prefill keeps every restore point it crossed, so retries resume where they left off instead of starting over. Resumed prefills no longer record restore points that don’t actually cover what they claim—on models with recurrent layers, this could force a request matching 46,000 of 47,000 tokens to reprocess from scratch. The update also disables Claude Code’s “tokens left” countdown system message, which Ollama had been moving to the front of the prompt and breaking the KV cache on every request.
Other improvements include a fallback to npx for the DeepSeek Harness launcher when a global npm install fails, with proper Windows command-shim support. The onboarding flow has clearer introductory text, a macOS header aligned with the native traffic-light buttons, and Cmd/Ctrl zoom shortcuts disabled during onboarding so the fixed window keeps its intended size. The MLX dependency has been updated, and default packaging has been fixed after macOS-specific assumptions were breaking Linux and Windows builds. Finally, the Apps header no longer overlaps the macOS traffic lights during sidebar open transitions—the header padding animation is now synchronized with the sidebar width animation.
This release candidate is a step toward the next stable Ollama version, following v0.32.15.
