Player not loading? Watch on YouTube
This tutorial walks through the speaker's Ollama HA app, which runs Ollama on the same machine as Home Assistant. The official Ollama integration connects the engine to conversation agents and AI tasks within automations. The speaker says they use this setup daily for forecast advice and camera snapshot descriptions.
The installation requires Home Assistant OS, a stable network connection, and storage for downloaded models. Steps include adding the custom repository, installing the app, checking that Ollama listens on port 11434, and configuring the integration. GPU support depends on supported passthrough. Local models need no additional account; cloud access requires signing in to Ollama.
To test a local LLM, the speaker recommends a small model of about 400 MB before trying larger downloads. They also demonstrate GPT-OSS-120B through Ollama's cloud service. Local inference uses the host's resources, while cloud prompts travel over the internet. Their preference for cloud models on weaker hardware is advice based on this setup.
Device control through Assist requires a model with tool support. The speaker recommends a backup and exposing only a few entities initially. Later examples use forecast data for clothing advice and a vision-capable model for camera notifications. The companion PDF contains YAML examples and troubleshooting; obtaining it also subscribes the requester to a newsletter.
