Teach your camera to notice anything.
Describe what you want it to catch and add a few example photos, and it trains in your browser. When it sees it, it can speak, show a notification, or send a message to Slack, Discord or a webhook.
Object detection runs in this tab, and the video isn't uploaded.
This demo recognises about 80 everyday objects without any training. For something more specific, like a particular mug or gesture, the builder trains a detector from your own photos.
How it works
Building a detector takes four steps and doesn't involve writing code or collecting a dataset.
Say what to notice
Describe it in your own words, or start from an idea such as “tell me when I bring a drink to my desk” or “tell me when someone walks up behind me”.
Show it a few examples
Take some photos with your webcam or upload them. Claude can check them and flag any that look like they're in the wrong group.
Train it in the browser
A small model is trained on your examples in the page. It usually takes a few seconds, and your photos aren't sent to a server.
Pick what happens next
When it sees what you trained it for, it can speak a phrase, show a notification, message Slack or Discord, call a webhook, or send an email.
Don't know what to build? Start here
Pick one of these to open the builder with the description and photo groups filled in, or write your own.
About you
- When I bring a drink to my desk
- When I've been slouching
- When I'm on my phone instead of working
- When I put my headphones on
Home
- When someone walks into the room behind me
- When the dog jumps on the couch
- When a delivery shows up at the door
Office & focus
- When someone walks up behind me
- When I've left my desk for a while
Whatever you want
- Start from a blank page
Pick what happens when it triggers
A detector can run one action or several at the same time.
Speak out loud
Speaks a phrase you write, such as “Why thank you, I'll enjoy that tea!”, when it triggers.
On-screen banner
Shows an alert on the page. There's nothing to set up.
Push notification
Shows a desktop notification, including while the tab is in the background.
Slack message
Posts to a channel through an incoming webhook.
Discord message
Posts to a Discord channel through a channel webhook.
Any webhook
Sends a JSON POST to a URL you choose, which works with tools like Zapier, Home Assistant and n8n.
Sends an email when the detector sees what it's watching for.
Frequently asked questions
Is Lookout free to use?
Training and detection run in your browser and don't cost anything. AI-assisted labelling runs on the site owner's Anthropic key, so it needs a free account and is capped at 600 photos a month. Email triggers depend on keys the site owner configures, and anything that isn't set up is disabled with a clear note rather than failing quietly.
Do I need to write code or collect a dataset?
No. You describe what to notice in plain English and add a handful of example photos from your webcam or your files. Lookout trains a small model on them in the page.
Do my photos get uploaded anywhere?
Training happens on your device and your photos aren't sent to a server for it. They're only uploaded if you turn on AI-assisted labelling for a photo, or create an account to back your detector up.
Do I need an account?
Not to build or run a detector. Detectors are saved in your browser as you work. An account is needed for two things: backing a detector up so you can restore it on another machine, and AI-assisted labelling, which spends the site owner's Anthropic key and so comes with a monthly allowance.
What can a detector do when it sees what it's trained for?
It can speak a phrase, show an on-screen banner, raise a browser notification, message Slack or Discord, call a webhook, or send an email.
Can I use a detector outside Lookout?
Yes. The export option gives you a zip with the trained model, the training photos, and a plain index.html that runs the detector on its own, so it can be used in your own project.