Generative AI without semantic search

[dpvu]

Describe what you are trying to accomplish and why in non technical terms

While semantic search is great, the RAM requirements exceeds available memory on devices like a Synology NAS or Raspberry Pi. The Gen AI descriptions are still useful so I'd like to be able to generate them and save them in Frigate (and have the MQTT description update) without turning on semantic search.

Describe the solution you'd like
Allow generative AI to populate the description without running a semantic search.

Describe alternatives you've considered
Upgrading RAM in my Synology NAS would be expensive. There is a Home Assistant LLM Vision add-on but the developer is considering deprecating Frigate support due to the Frigate built-in gen AI capabilities but this requires semantic search.

Additional context
Semantic search does work with only 4 GB of RAM but after ~24 hours detection will no longer work until Frigate is restarted.

[NickM-27]

semantic search is required for generative ai as the embeddings are needed to be able to allow searching through the generative ai descriptions

[dpvu]

semantic search is required for generative ai as the embeddings are needed to be able to allow searching through the generative ai descriptions

I don't really need to search in them. Mostly just trying to generate a description for a HA notification.

[hawkeye217]

I don't really need to search in them. Mostly just trying to generate a description for a HA notification.

You'd likely have better luck with using Generative AI in Home Assistant, then.

https://www.home-assistant.io/integrations/google_generative_ai_conversation/