There is one specification that sorts almost every doorbell into the right pile, and it is rarely on the front of the box: does it expose a standard video stream on your own network?
Look for RTSP, or ONVIF which usually implies it. If a doorbell publishes an RTSP stream, any recording software on your network can pull it, your automation platform can display it, and none of that depends on the manufacturer. If it does not, the video only exists inside their app and their cloud, and every integration you find will be a workaround.
That single question separates the market more cleanly than any amount of review reading. Manufacturers that support it usually say so in the technical specifications, sometimes buried, sometimes only in a firmware release note. If you cannot find it stated anywhere, assume it is absent.
The other three things to check, in order:
Where the button press event goes. A doorbell has two jobs and they fail independently. You can have a perfect local video stream and still have the press only arrive through the vendor's push notification. Look for the press being exposed as something on your own network — a webhook it can call, a local API, or support for one of the local protocols. Without that, your automations cannot react to somebody being at the door, which is usually the actual reason you wanted this.
Whether it works with no internet. Test this on day one, not eventually: pull the internet connection, keep the local network up, and see what still functions. Many devices need cloud reachability even for local features because the app calls home to authorise the connection. That is the failure you will care about during an outage.
Power. This is the constraint people discover last and it decides everything else. Battery doorbells sleep to save power, which means they cannot hold a continuous stream, which means no continuous recording and a noticeable delay from press to picture. A wired doorbell — mains through a transformer, or power over ethernet — can stream continuously. If you want reliable local recording and low latency, you want a wired one. No amount of software fixes a device that is asleep.
What genuinely does not work locally regardless of what you buy:
- Pre-roll, the few seconds before the trigger. Requires the camera to be continuously recording, so battery models cannot do it.
- Remote access from outside the house without either exposing something or running your own tunnel or VPN. Doable, and it is work you now own.
- Vendor-specific cleverness — package detection, familiar-face recognition — is done on their servers. Local equivalents exist and run on your own hardware, but they are a separate project.
- Two-way audio is the single most commonly broken feature in local integrations. It frequently works in the vendor app and not through anything else, even when video is fine. If it matters to you, verify it specifically rather than assuming it comes with the stream.
The practical shape that works: a wired camera with RTSP and a local event, feeding recording software on a machine you own, with your automation platform subscribing to the events. Set up once, no subscription, and the manufacturer cannot change the terms.