🖼️ Home Assistant – LLM Vision

Home Assistant LLM Vision integration: analyze camera images with AI, query the vision timeline, and create timeline events.

Source: openclaw-hub

llmvision_get_image

Download a keyframe image from HA LLM Vision media storage. Pass a key_frame path from a timeline event (e.g. /media/llmvision/snapshots/xxx.jpg). Returns the local file path for use with the image or message tools.

Name	Type	Required	Description
key_frame	string	Required	The key_frame path from a timeline event (e.g. /media/llmvision/snapshots/abc123-camera0.jpg).

llmvision_timeline_get

Get events from the LLM Vision timeline (calendar.llm_vision_timeline). Returns a list of AI-generated observation events with timestamps, summaries, and descriptions. Useful for reviewing what the cameras have seen recently.

Name	Type	Required	Description
start_time	string	Optional	Start of the query window as ISO 8601 (e.g. ‘2026-04-09T00:00:00+00:00’). Defaults to now minus days.
end_time	string	Optional	End of the query window as ISO 8601. Defaults to now.
days	integer	Optional	Number of days to look back when start_time is not set (default: 7).
limit	integer	Optional	Maximum number of events to return (default: 50, max: 200).

llmvision_analyze_image

Trigger an AI image analysis on a Home Assistant camera entity using LLM Vision. Sends the current camera snapshot to the specified AI provider with a custom prompt and returns the AI-generated description. Can optionally store the result in the timeline.

Name	Type	Required	Description
camera_entity	string	Required	Camera entity ID to analyze (e.g. camera.front_door).
message	string	Required	Prompt / question to send to the AI about the image.
provider	string	Required	LLM Vision provider to use (e.g. ‘anthropic’, ‘openai’, ‘ollama’).
model	string	Optional	Specific model override (optional, uses provider default if omitted).
store_in_timeline	boolean	Optional	Whether to save the result as a timeline event (default: false).
expose_images	boolean	Optional	Whether to expose the captured image in the timeline event.
generate_title	boolean	Optional	Whether to auto-generate a title for the timeline event.
response_format	string	Optional	Response format from the AI: ‘text’ (default) or ‘json’. Allowed: text, json.
max_tokens	integer	Optional	Maximum tokens for the AI response.

llmvision_create_event

Create a new event in the LLM Vision timeline. Use this to manually log observations or detections with optional camera image, label category, and time range.