ChatGPT in your camera: the complete guide

From “I wish someone was watching that” to an Agent that is. Everything you need to set it up, write Events that work, and trust what you get back.

The idea in one sentence

You already know how to ask AI for help in ordinary words. Now imagine asking a camera: “Tell me if anyone climbs the fence.” And it just… does. Tonight, tomorrow night, and every night after, without you asking again.

That’s what people mean by “ChatGPT in your camera”. To be precise: Agentinosaurus is an independent product that uses OpenAI models by default to look at camera frames against the instructions you write. You manage everything in Agentinosaurus, and you use your own OpenAI API key for the analysis.

The difference from a chat is the part that changes your life. A chat answers once. An Agent keeps watching.

Before you start: pick the view that keeps you up at night

Don’t start with your most impressive camera. Start with the one that covers the thing you actually worry about. The front door where parcels vanish. The back entrance nobody uses after six. The hallway an elderly parent walks at night. An alert there will mean something the first time it arrives.

You’ll need three things: a camera on your network, a Windows or Linux computer on the same network that can stay on, and an OpenAI API key with API billing enabled. A ChatGPT subscription doesn’t include API usage: API billing is set up separately on the OpenAI platform, and it only takes a few minutes.

Light and angle matter more than resolution. A wide shot tells you what’s happening in a space; a closer shot lets the Agent see smaller actions. Point the camera at the job, not at the view.

Step 1: Pair your Station

Station is the small app that connects your cameras to your workspace and keeps the watch going. Download it, install it, and follow the pairing steps. Give it a name you’ll recognize at a glance, like Home or Main shop.

Then add your OpenAI key in Settings → Stations. Your Station uses that key to send frames directly to OpenAI. Leave the computer awake and online: you can close the browser, and the watch carries on.

Step 2: Write your first Event

Write it the way you’d brief a person standing next to you. “Someone climbs over the garden fence.” “A box is left in front of the emergency exit.” “A person falls and stays on the floor.” Short, visible, specific.

The best Events describe what the camera can actually see: an action, an object, a place. “A person tries the handles of parked cars” works far better than “something suspicious happens”, because the Agent knows exactly what to look for, and you know exactly what an alert means.

Your first wording is a draft, not a contract. Read what comes back, sharpen the sentence, and add new Events whenever a new worry shows up. One camera can carry several.

Step 3: Add your camera and choose how it watches

Choose your Station and let camera detection find your device. If it doesn’t appear, the manual connection guide shows you how to enter its address. Name the camera after the place, not the model, and attach your Event.

Then decide how this camera should watch. “When movement is detected” focuses analysis on activity and still checks quiet scenes from time to time. “Continuously” analyzes every window, including moments where nothing seems to move, and uses more AI tokens.

Make that choice per camera. A busy entrance might be fine watching for movement. A view where stillness is the problem, like someone on the floor, a door left open or a box that shouldn’t be there, may deserve continuous attention.

Step 4: Test it like you mean it

Act out a safe version of the moment. Leave a parcel by the door. Walk into the stockroom. Sit down on the hallway floor for a minute. Then read the Message your Agent sends. Does it say what you’d want to hear, at the moment you’d want to hear it?

Turn on Push Notifications on every phone or computer that should hear about incidents; Messages also stay in your workspace for later. On Business, Webhooks can send incidents straight into the tools your team already uses.

Watch the monitoring status too, not just the alerts. An Agent can only keep watch while its Station, its camera connection and the internet are up, and you’ll be told when that watch is interrupted.

What it costs, and where your images go

Your Agentinosaurus subscription covers the product and your camera slots. Your selected AI provider bills analysis to your own account. Our calculator shows how cameras, hours, activity and the number of Events change that number, so there are no surprises.

Frames travel straight from your Station to your selected AI provider, with OpenAI as the default. They don’t pass through our servers. We store the results and Messages; incident snapshots stay on your Station.

One honest note: an Agent is a very attentive helper, not a guarantee. It can misread a dark scene or miss what the camera can’t see. Keep your smoke alarms, locks and emergency services exactly where they are, and let your Agent be the extra pair of eyes on top.

From one camera to a quiet mind

Get one Event right on one camera, and live with it for a few days. You’ll notice the moment it stops being a gadget: the first time your phone buzzes with something you’d genuinely have wanted to know, and you didn’t have to be looking.

Then add the next view, and the next worry. The goal isn’t more alerts. It’s the right ones, around the clock, about the things you actually care about.