Recognition

Recognition lets you hand the AI a picture and ask “what am I looking at?” It reads the photo the way a person would: the overall scene, the things in it, any text or signs, vehicles down to the make and model, and anything that looks like a safety concern. It all runs on a vision model on your own server, so the photo never leaves your hardware.

Analyzing a Photo

Open the Recognition tab in the Images app, drop in a photo, and the AI works through it in a few passes at once. When it’s done you get a tidy panel:

• Scene: a plain-language description of what’s going on and where.
• Objects: the things it spotted, each with a confidence bar.
• Vehicles: type, color, brand and model where it can tell (a Tesla Model 3, an F-150), and a plate if one is readable.
• Text: anything written in the image — a sign, a label, a document — with the language it’s in.
• Safety: a flag for anything worth a second look, like fire, a weapon, or a hazard, marked from normal to critical.

You can pick which of these you care about before you run it, and every analysis is kept in a history gallery so you can go back to it later.

In Chat

You don’t have to open Recognition to use it. In a conversation with your companion, attach a photo and just ask about it — “what breed is this?”, “what does this label say?”, “is that a package on the porch or a person?” The companion looks at the image with the same local vision model and answers in the flow of the chat.

For Peace of Mind

Recognition is genuinely useful for keeping an eye on things: glance at a camera snapshot and know whether that’s a delivery or a stranger, read the fine print on a medication box, or identify an appliance you can’t name. Because it runs at home, you can point it at family photos and doorstep snapshots without a single one being uploaded, profiled, or stored on someone else’s server.

A Note on the Internet

Recognition runs on a vision model installed on your own server. Once that model is in place it works completely offline, and no image you analyze ever leaves your network. A larger vision model sees more detail — worth choosing if you lean on the safety flags.