Nobody should have to watch the footage.
Upload the recordings from every camera. Search for a person, a vehicle, a plate or a voice — by photo or in plain words. VidForgeX shows you exactly which file and which second to look at, ranked and explained. Your investigator makes every call.
The problem
Cameras are everywhere. Watching them back is still a person, a chair and a scrub bar.
One incident on a site with 16 cameras means days of recordings. A store detective, a police investigator, an airport security team — all of them lose hours or days scrubbing footage to find a few seconds that matter. Often they give up before they find it.
16 cameras × 3 days, watched at normal speed
1,152 hours
What the case actually needs: 4 moments, a few seconds each.
Search
Ask for what you're looking for. Across every camera at once.
One search box. A photo, a crop from another video, or a few words. Results come back as strong, possible, or worth a look — never hidden, always ranked.
Click a mode to try it.
Reference photo
cropped from the entrance camera
Entrance
0:14:03
Strong matchAisle 6
0:18:45
Strong matchCar park A
0:31:02
Possible matchSee an investigation unfold
From surveillance footage to an investigator-reviewed investigative trail.
Where did the white SUV go after leaving the North Gate?
3 uploaded recordings · 2 live cameras · 1 reference vehicle
Or select a stage above to explore at your own pace.
What happened
It tells you where to look before you even ask.
Every recording is screened frame by frame. The moments worth a human's attention come to the top of the page, each with a timestamp you can click.
- Bags and objects left behind — how long they sat alone, and who picked them up.
- Watchlist alerts — add someone once and every new upload is checked for them, and every old one too.
- Plate hotlists — get told when a listed plate is read.
- Vehicle speed — mark a known stretch of road once per camera and get measured speeds.
- Crowd size — estimate how many people are in a dense scene.
- A written account — an AI draft of what the recording shows, for the investigator to check and edit.
- 0:14Person enters with a backpack
- 0:55Backpack left alone
- 1:06Picked up by a second person — linked to them
- 1:07Watchlist: possible match to a listed person
- 1:31Silver hatchback leaves · plate read
A voice heard once can be looked for everywhere. Upload a short clip and get back the segments across analysed footage where a similar voice occurs — speaker-level acoustics, not word matching, so it works in any language and needs no transcript.
Evidence you can stand behind
Built for the day someone asks, “How do you know?”
An investigation tool is only as good as its answers under questioning. Every result carries its own paper trail.
Fingerprint check · lobby-cam1.mp4
- Fingerprinted on arrival
- Each file gets a SHA-256 fingerprint at upload and again at analysis. If they differ, the report says so.
- Decisions are permanent
- Every “yes, that’s him” or “no, it isn’t” is recorded with who and when. Changes are added, never overwritten.
- The method is named
- Every result states which AI models produced it — the first question a defence expert asks.
- Gaps are shown, not hidden
- If part of the analysis could not run, the page says so. “Nothing found” never means “not checked”.
- Case files and export
- Pin moments to a case, build the timeline, and export a versioned record that cannot be edited afterwards.
- A human makes every call
- The system never identifies anyone on its own. It points; your investigator decides — and that decision is logged.
Measured, not promised
We publish the numbers — and the conditions behind them.
Tested on real CCTV and public benchmarks, not a hand-picked demo.
0.935
Vehicle matching score (AUC)
Same vehicle vs different vehicle across cameras; 0.951 on our own CCTV, up from 0.63 on our previous engine. 1.0 is perfect.
Tested on: VeRi-776 public benchmark · 2,644 images, 180 vehicles
22 / 22
Licence plates read exactly
Barrier CCTV clips with known plates; cameras close and head-on.
Tested on: AGH University of Kraków Parking Database
≤ 12%
Speed vs GPS ground truth
Pole camera with a marked stretch of road, against an RTK-GNSS-tracked car. Within 5% at peak in our lab test.
Tested on: Intersection Monitoring dataset (CSIC)
+4.5%
Cost to screen every frame
Motion screening checks every frame and sends only the interesting ones to the heavy models.
Tested on: Measured on our own footage
What we don't claim yet — and say out loud
A system that points you to the wrong place costs a glance. One that stays quiet when it shouldn't costs the case. We tune for the first.
Who it's for
The same pain, in every building with cameras.
- Retail loss prevention
- Follow a subject from entrance to exit across the whole store.
- Investigators & detectives
- Turn a case’s footage into a timeline without scrubbing it by hand.
- Airports & transport
- Find the owner of a left bag, or where a person went next.
- Security operations centres
- Watchlists and plate hotlists across the camera estate.
- Campuses & facilities
- Answer “who was here, and when?” without a week of review.
How it works
From a hard drive of footage to a case file.
- 01
Upload the footage
Any common video format, from any camera or DVR export. Tag which camera it came from.
- 02
Every frame is scanned
People, vehicles, plates, objects and voices are indexed so they can be searched.
- 03
Search and review
Ask for what you need. Confirm or reject each result — every decision is logged.
- 04
Build the case
Pin the moments, see the timeline, export a fingerprinted record.
Built for uploaded footage first. Live camera connections with watchlist and plate alerts are in pilot.
Stop watching. Start finding.
Bring footage from a real case to a demo and we'll search it with you — person, vehicle, plate, and the evidence export at the end.
Access by application
Forensic is reviewed before it is switched on.
Footage of people is sensitive, so every workspace is approved by our team first. Tell us who you are and what you investigate — we reply within about two business days.
- 01
Apply
Who you are and what you investigate.
- 02
Review
Our team checks every application, usually within two business days.
- 03
Onboard
Your workspace is switched on and we walk you through a first case.
Already approved?
Explore
- Forensic video intelligenceHow AI turns hours of CCTV into a few moments to review.
- CCTV video analyticsPeople, vehicles, plates and events — searchable.
- Person tracking across camerasFollow one person through every camera, with a human deciding.
- Crowd analyticsEstimate how many people are in a dense scene.
Nobody should have to watch the footage.