ha-llmvision

Why we included this project

For Home Assistant users with cameras, ha-llmvision puts a multimodal LLM on top of your existing video feeds and Frigate events, so the system can describe what's happening, answer questions about a clip, and feed extracted details into sensors. It keeps a timeline of camera events and remembers people, pets, and objects across them, which you can surface on a dashboard or ask about through Assist. The integration works with a broad set of providers, from OpenAI, Anthropic, and Gemini down to local options like Ollama and LocalAI, letting you pick between cloud convenience and full privacy. That range makes it a straightforward way to get more value out of hardware you already own, whether your goal is richer security alerts or automated scene descriptions.