LLM Vision

LLM Vision is a Home Assistant integration for people who want camera footage to inform smart home alerts and automations. It uses multimodal models to describe what cameras capture and extract data from images.

• Camera and video analysis: Ask questions about images, local video files, live camera streams and Frigate events. It can use any camera stream viewable in Home Assistant.
• Event notifications: Add model-generated descriptions to camera and Frigate alerts. For video, the integration selects relevant frames and a representative image for the notification.
• Event history: Optionally save analyzed events in a dashboard timeline and Home Assistant calendar entities. You can limit saved events to those classified as important, or ask Assist about past activity with a conversation agent such as Extended OpenAI Conversation.
• Entity updates: Extract data from a camera image, image entity or local image file to update number, text, selection or boolean entities. This action doesn't accept video.

The integration is free and licensed under Apache 2.0. It runs in Home Assistant and works with local model providers such as Ollama and LocalAI, as well as Open WebUI, OpenAI, Anthropic, Google and providers with an OpenAI-compatible API. Local providers can keep image analysis on your hardware; cloud providers process inputs through their services. Saved event records remain in your Home Assistant instance after analysis.