Connect
Point OmniVision at your cameras, recorders and archives. Live RTSP and ONVIF streams, recorded files, dashcam footage and cloud buckets all go into one index.
Video intelligence for the cameras you already own
Ask your cameras anything.
OmniVision watches every camera and every archive, writes down what happens, and lets your team search it, chat with it and get alerted about it in plain English. It runs on a box in your building, so your footage never has to leave it.
Works with RTSP & ONVIF cameras
Runs on your premises
Windows or Linux
LIVECAM 04 · North gate22:41:08
person · 0.94
package · 0.88
car · 0.91
RTSP & ONVIF cameras
Ring doorbells & cameras
Dashcams with GPS
S3, Azure & Google Cloud archives
Local AI, no cloud required
Signed evidence exports
Append-only audit trail
Windows or Linux
How it works
From hours of footage to answers in seconds
No new cameras and no cloud upload. OmniVision sits beside what you already have and turns it into something you can talk to.
Understand
On the edge box, it detects people and vehicles, and a vision-language model writes a plain-English description of each scene: who, what, carrying what, and where.
Ask
Search the way you would ask a colleague, chat with your footage and get answers with clips attached, and write alerts as sentences instead of configuring zones and schedules.
The platform
One engine. Seven ways to use it.
Find it the way you'd describe it
"Anyone wearing pink in the office?" "A white van at the loading bay last Tuesday." Time of day is a real filter, not a guess.
Chat with your footage
Ask follow-up questions and get written answers. Every claim cites the moment it came from, so you can watch it.
Write the rule as a sentence
Every rule shows how it was understood before it ever fires.
Count what crosses the line
Draw a line, get every vehicle counted once, with its type and direction.
Proof that stands up
Signed exports with hash manifests, and an append-only audit trail behind every search, report and export.
Where, when and how fast
Routes on a street map stored on the box, plus speed and driving events (braking, cornering, speeding), all from the camera's own GPS. No extra hardware on the vehicle.
A watchtower, not a wall of screens
Live view, incidents that open themselves, a morning digest of the last 24 hours, and push alerts to phones and browsers.
Ask OmniVision
Ask it like you'd ask the night guard
Pick a question to see the kind of answer OmniVision writes: short, specific, and with every claim tied to a clip you can open.
Illustrative example conversations. Real answers come from your own indexed footage.
Yes, twice. A person in a dark hoodie tried the back door at 22:14 S1 and walked off towards the car park at 22:16 S2.
At 23:41 a delivery driver left a parcel by the door and drove away S3.
S122:14 · Back door
S222:16 · Car park
S323:41 · Back door
47 vehicles crossed the gate line today C1: 39 cars, 6 trucks and 2 motorcycles.
The busiest hour was 08:00 to 09:00, with 14 vehicles, 12 of them coming in.
C1Gate line · today
08:12 · Gate
08:41 · Gate
Quiet until 07:30, when the first truck backed into bay 2 S1. Two people unloaded pallets until about 08:10 S2.
A second delivery arrived at 09:45 S3; the bay was empty again by 10:20.
S107:30 · Bay 2
S207:52 · Bay 2
S309:45 · Bay 1
Not as far as the footage shows. Bags appeared in reception six times today, and each time the owner stayed with it or took it away within a few minutes S1 S2.
OmniVision looks at sampled moments, not every second, so open the clips if it matters.
S111:05 · Reception
S215:32 · Reception
15:36 · Reception
Private by design
Your footage stays in your building
Detection, captions, search and the AI that answers questions all run on the OmniVision box at your site. Nothing is uploaded unless you choose a cloud-assisted option, and even then only still images leave, never video.
- Local language and vision models: no third-party AI service sees your cameras
- Remote access through an encrypted tunnel, with no ports opened on your router
- Roles for admins, supervisors, guards and auditors, with every action logged
Your camerasExisting IP cameras and recorders
OmniVision EdgeBoxDetection, captions, search, chat, recordings
Your teamBrowser, tablet or phone
Your premises
Remote accessOptional, through an encrypted tunnel
Solutions
Same engine, different floor
Security is the first job, not the only one. Anywhere cameras already watch, OmniVision can answer questions about what they saw.
Facilities & campus
Offices, schools and estates: after-hours access, lost property, visitor questions.
"Who used the side entrance on Saturday?"Retail
Queues, dwell time and shelf gaps for store operations, not only loss prevention.
"When was the checkout queue longest today?"Construction
PPE and exclusion zones in plain English, with signed evidence for insurers.
"Anyone in the crane zone without a hard hat?"Warehouse & logistics
Dock dwell, gate traffic and near-misses, counted and searchable.
"How long did trucks wait at bay 3 yesterday?"Fleet & insurance
An incident is reported at a time: which clip, where, and how fast.
"Show the van's footage near the junction at 14:20."Assisted living
Falls and wandering, with privacy-aware summaries written for families and staff.
"Did anyone leave the east wing overnight?"Hardware
Meet the EdgeBox
One quiet box per site runs everything, from detection to chat, and powers up to four cameras directly. Battery backup for power cuts is in development.
Camera ports4 × PoE+ built in
AI on board8, 16 or 24 GB GPU
Power cutsBattery backup in development
RunsWindows or Linux
OmniVision
Questions
Good to know
Do we need new cameras?
Usually not. OmniVision works with most IP cameras and recorders that provide an RTSP stream or support ONVIF, plus Ring cameras, dashcam footage and recorded archives.
Does our footage leave the site?
Not with the standard setup: detection, captions, search and Ask OmniVision all run on the box at your site. A cloud-assisted option exists for sites without GPU hardware; it sends still images of detected scenes, never video, and only if you choose it.
How accurate is it?
It depends on camera placement, light and resolution. OmniVision looks at sampled moments rather than every second, and every answer links to the clips behind it so a person can check. During setup we test it on your own cameras before you rely on it.
What hardware do we need?
One EdgeBox per site, or a compatible mini PC. Captions and chat run best with an NVIDIA GPU of 12 GB or more; everything except the AI writing works without one. The EdgeBox page has the details.
How do we get started?
Create an account and answer a short questionnaire about your cameras and network. You can save it and come back, and invite your IT contact to answer the technical parts. We use it to recommend the right setup and prepare a quote.
Get started
Tell us about your site. We'll do the rest.
A guided questionnaire about your cameras and network. Save as you go, invite colleagues to answer the technical parts, and we'll come back with a recommended setup and a quote.
About 40 questions · "Don't know" is always a fine answer