Ollama, the local AI framework, released version 0.19.0 today, March 30, 2026. This update introduces a new web search plugin for its 'ollama launch pi' command, enabling direct web search capabilities for local AI models. The release also focuses on performance and stability enhancements for the MLX runner, including improved KV cache hit rates for Anthropic-compatible APIs and fixes for KV cache snapshot memory leaks. Additionally, specific issues were addressed, such as incorrect "model is out of date" warnings, tool call parsing for Qwen3.5, and flash attention activation for grok models. The update also resolves a loading problem for the qwen3-next:80b model. This release continues to refine the local AI experience, expanding functionality and reliability for developers and users.