User wants Frigate doorbell notifications to be processed through an LLM to generate more natural, personalized messages (e.g. "David's at the door with the dog 馃悤!") instead of standardized alerts. Need to scope out required HA integrations: openai_conversation or ollama integration, face recognition setup, known object/person labeling, and time-of-day context. Steps to investigate: 1) Which LLM integration to use (OpenAI vs local Ollama), 2) Face recognition options (double-take, compreface), 3) How to pass Frigate event data to the LLM in an HA automation, 4) How to route the LLM response back to push notifications.
User wants Frigate doorbell notifications to be processed through an LLM to generate more natural, personalized messages (e.g. "David's at the door with the dog 馃悤!") instead of standardized alerts. Need to scope out required HA integrations: openai_conversation or ollama integration, face recognition setup, known object/person labeling, and time-of-day context. Steps to investigate: 1) Which LLM integration to use (OpenAI vs local Ollama), 2) Face recognition options (double-take, compreface), 3) How to pass Frigate event data to the LLM in an HA automation, 4) How to route the LLM response back to push notifications.