LLM Vision

LLM Vision is a free, open-source Home Assistant integration for people who want their smart home to interpret what cameras see. It uses multimodal LLMs to describe images, video files, live feeds and Frigate events, then uses those results in notifications and automations.

You choose where analysis runs. Ollama and LocalAI let you use models on your own hardware; hosted providers such as OpenAI, Anthropic and Google process inputs through their cloud services. It also supports Open WebUI and providers with an OpenAI-compatible API. The integration uses the Apache 2.0 license.

Camera events can become a searchable history of activity around your home. Event storage is optional, and analyzed events stay in your Home Assistant instance. You can retain only events the model classifies as important, view them in a dashboard timeline, or ask about past activity through Assist with a conversation agent such as Extended OpenAI Conversation. Events also appear as calendar entities for automations.

For live feeds and video, the image pipeline selects relevant frames and a key image for notifications. It supports hardware acceleration and reduces image resolution to limit latency and token use. Any camera stream viewable in Home Assistant can serve as an input.

Image analysis can also extract values and update Home Assistant entities, including numbers, text, selections and booleans. This data extraction accepts cameras, image entities and local image files; it doesn't accept video.