Yes. Look into Frigate.
The main qualifier is that you need to convert your Wyze stream into RTSP (or similar protocol) and then you can feed that into Frigate. It not only has native object detections built in such as per person, package, and many animals, but it now also includes a generative AI module. You can aim it at your local AI model for processing.
You can then even configure what the AI considers important. Frigate uses an activity_context_prompt in your config.yml. You define logic in plain English:
- Level 0 (Normal): “A person walking a dog during the day,” “The mailman dropping off a letter.”
- Level 1 (Alert): “Someone looking into car windows,” “A person in the backyard after 11 PM,” “An unidentified person lingering for more than 30 seconds.”
Combine that with Home Assistant. Then the AI processes the snapshot or clip, compares it against your prompt, and returns a “score” or “importance” level. You can then tell Home Assistant to only send a notification if importance == 1.
If you already have a local model, you can even use CLI or openclaw to give the local model “hands” to help set this up for you.
Warning - only use CLI or openclaw on a separate sandboxed system with the proper precautions set up including backups, white listing, human in the loop, etc.
In fact, you could bypass frigate entirely. You could set up an rtsp stream to simply take a thumbnail image at a set interval. And copy it into a read-only folder that open claw has access to, and it could check the image itself and report back to you if an image contains something similar to the instructions you gave it.
There are definitely a lot of ways you can do this that would work using a local AI.
Again, step 1 is RTSP. Depending on what model camera you are dealing with, there are various options to achieve this. Some models currently have beta firmware that gives rtsp natively now.
For some camera models, you can use the Thingino firmware To convert the camera to a local only rtsp cam. For a lot of the other cameras you might be able to use something like docker Wyze bridge to convert the Wyze stream to RTSP. But some of the cameras have received newer firmware which seems to be interfering with that, so you might have to roll back the firmware to an earlier version for that option to work.
One of the easiest options that work with pretty much all of the cameras is to use tinycam Pro in server mode (Either using an old Android phone that you just leave at home, or on an Android emulator on a computer that you just leave running), and then you have all cameras with rtsp.
There are a few other options, but those are the most popular.
Hope that helps.