How to search through hours of CCTV footage

Anyone who's had to find something on CCTV knows the routine. Something happened on Tuesday, probably in the afternoon, and now someone has to sit at a monitor scrubbing at eight times speed until their eyes give out. An hour of video takes ten minutes to skim badly, and a shop with eight cameras records close to two hundred hours a day. The maths never works in your favour.

Why scrubbing doesn't scale

Traditional video systems store pictures, not meaning. The recorder knows the camera and the timestamp and nothing else about what's in the frame, so the only search tool you have is your own attention. Motion detection narrows things a little, but on a busy site everything is motion. That's why footage mostly gets reviewed after something has already gone wrong, and why most of it is never watched at all.

Make the footage searchable first

The fix is to have the footage read before you need it. Svid pulls frames from any video source and has a vision model describe what's happening in each one: people, objects, actions, hazards. Those descriptions are stored and embedded for semantic search, so finding a moment stops being a viewing problem and becomes a search problem. Search problems take seconds.

1. 01

   Create an account

   Sign up at app.svid.ai. Onboarding asks four questions (business type, what to watch for, camera count, hours a day) and shows you a price with the working on screen: based on your own usage, with a $9 a month minimum.

2. 02

   Add the footage

   The Add tab takes an uploaded file, a pasted video URL (the title fills itself in), a live phone camera at app.svid.ai/live.html once you've signed in on the device, or an RTSP feed from an IP camera, captured on our side. Last week's export works as well as this morning's stream.

3. 03

   Let it read every frame

   Frames are sampled at your watch level: Precise looks every second at 1080p, Standard every 5 seconds at 720p (the default), and Economy every 15 seconds at 480p. A vision model writes a plain-English description of each kept frame, and frames where nothing moved get no AI calls.

4. 04

   Type what you're looking for

   Footage sits in a Library of nested folders, and the search bar searches the folder you're in, subfolders included. Search in ordinary language: “person in a red jacket near the till”, “someone slipped in the entrance”. You get matching moments with timestamps, not a wall of thumbnails.

What you can search for

• People, by description: “man in a high-vis vest and black boots”
• Actions: “someone putting an item in their pocket”, “person climbing the fence”
• Vehicles: “white van at the loading dock after 6pm”
• Situations: “queue longer than four people”, “child left the play area alone”

What it costs and where it runs

Search isn't an extra; it's the core of the product. The monthly price is based on your own usage for your cameras, floored at $9 a month, and the whole derivation is on screen before you pay: frames per month, AI analyses, and your price. If you also want Svid to raise alerts or write reports, each check is priced in the Store as the extra it would add at your own usage. Frames are described in the cloud by a vision model; for sites that can't allow that, a fully on-prem appliance mode runs local models and nothing leaves the network. The detail is in the pricing section.

How long does a search actually take?

Seconds. Once footage has been read, a search is a text query over stored descriptions, not a scan through video. Typing the sentence is usually the slow part.

Does it work with my existing cameras?

Yes. If it produces footage, Svid can read it: RTSP feeds captured on our side, exported files, a video URL, or a phone pointed at the scene. There's no proprietary hardware.

Can I search live footage too?

Yes. Live and recorded video go through the same pipeline, and live sessions are saved as videos, so this morning's monitoring is searchable this afternoon. Checks can raise alerts on the same feed if there are things you don't want to miss.

Do I need to train it on my site?

No. The vision model describes scenes out of the box. You connect footage and start typing.

What does searching my footage cost?

The price is based on your own usage for your cameras, with a $9 a month minimum, quoted at signup with the arithmetic shown. Search is included; checks that alert or write reports are priced separately in the Store from your own usage.