v0.20.5

Ollama, the open-source platform for running large language models locally, has released version 0.20.5 with significant new capabilities for deploying AI chatbots. The headline feature is the OpenClaw channel setup, which allows users to connect their locally-run Ollama models directly to popular messaging platforms including WhatsApp, Telegram, and Discord. This means developers can now build and deploy conversational AI agents that interact with users through these channels without needing complex middleware or cloud infrastructure.

Technically, the update also enables flash attention for the Gemma 4 model on compatible GPUs, which should provide noticeable performance improvements during inference by optimizing memory usage during attention computations. The release fixes the `/save` command for models using safetensors-based architectures and improves detection for curl-based OpenCode installations. These enhancements make Ollama an even more compelling tool for developers wanting to build and test AI applications with privacy and cost control by keeping everything local.