Camera, Frigate: Intelligent AI-powered notifications

Got it. I’ll add time and zone conditions as well as notifications to multiple devices soon!

As long as you have camera entities in home assistant that change their state or some binary_sensor entities that get triggered when something happens, this will still work.

Thank you so much for your response. So something like binary_sensor.XXXXX_nvr_driveway_motion would work?

Yes exactly. The automation will be triggered by the on state of a binary sensor. You can add multiple cameras and each camera can have its own binary sensor. Just add them in the same order.

Thank you. I will give this a whirl.

v1.3 of LLM Vision is out now and with it an updated blueprint.
Most notably frigate_url is no longer required. Instead previews are now fetched internally through the Frigate Integration.
For ‘Camera’ mode you can now also choose between a live preview and a snapshot of the actual event.

The biggest update to the integration is the new option to remember events. This adds a calendar entity, so you can easily see what’s been happening at a glance. It also enables asking about past events. You can read more about that here:
https://llm-vision.gitbook.io/getting-started/asking-about-events

This is step two of making your cameras smarter and more useful and follows last release, which rethought how camera notifications should work. The next step is to integrate LLM Vision even more into your dashboard. Stay tuned!

2 Likes

An issue with the new blueprint. When I select snapshot, I get no picture in the notification. But the image is processed and I do get an AI response.
One final issue. Is it possible to send the notification to multiple recipients? Now I can only choose my phone.

Hmm the new blueprint doesn’t fire when on Frigate mode. I have it on Frigate mode and set the frigate entities, but it doesn’t fire

Did you update the integration as well? There are some changes that enable including the snapshot in the notification. You also need to have the Frigate Integration istalled.

You can verify the snapshot was created by checking if the image is in /www/llmvision.

Sending the notification to multiple devices is something I’m working on right now!

I assume by Frigate entities you mean the cameras.
Have you renamed them or changed their ids? They need to have the same entity_id as in Frigate.
If no, could you check the traces of your automation? Are there any errors?

I wasn’t getting any errors and the entity IDs were the same. I got it working by using Camera mode. However, now I get duplicate notifications. The initial notification that has a picture and a motion detected caption, and then a second layer I get a notification with the LLM vision description

Are you on Android or iOS? It is possible that notifications need to be updated differently on android. I have only tested this using iOS.

Yeah I’m on android

I may have found the issue. Home Assistant docs mention unexpected behaviour when using the same tags across different groups on Android (Introduction | Home Assistant Companion Docs). I’ll try to fix this in the next update.

There are no images in that folder. Except some older ones. Do I need to use frigate, since I configured it to use the camera? Somehow frigate gives me no results. No motion or people detected.

v1.3 seems to have a few issues. If you’d like to test the blueprint beta, import this url https://github.com/valentinfrlch/ha-llmvision/blob/data-analyzer/blueprints/event_summary.yaml and overwrite the existing blueprint (your configuration will remain). If you encounter any issues please report them!

v1.3.1 also adds multi-device notifications.

Can’t import the blueprint.

Sorry this is the correct url:
https://github.com/valentinfrlch/ha-llmvision/blob/56ab80cef76b0f214757c3e6c0123423555d8113/blueprints/event_summary.yaml

I had accidentally deleted the dash.

That imported fine.I did a quick test only. I selected ‘camera’ again instead of frigate. The snapshot works a lot better. It first sends a notification, with a picture that motion is detected. Then after a few seconds it updates the picture to a few seconds later (much quicker then live preview) with the AI description. So that’s much better.

Nice blueprint, works fine for me.
I have a question about the tap navigate. Is it possible to show a big snapshot instead of going to a lovelace dashboard? I’m using the frigate mode.

Thx

1 Like