Omnisight

AI surveillance,
on the cameras you already have.

A real-time AI brain that sits on top of any RTSP-capable CCTV network. Detects threats, tracks vehicles and people across cameras, reads plates, and surfaces alerts in seconds. Built for command centers and enterprise security operations where the cost of missing something is real.

A real-time AI brain that sits on top of any RTSP-capable CCTV network. Detects threats, tracks vehicles and people across cameras, reads plates, and surfaces alerts in seconds. Built for public safety agencies and 911 dispatch centers where the cost of missing something is measured in lives.

<350ms
Median alert latency
42K+
Active cameras
62B
Daily detections
60+
Languages supported
DEPLOYED FOR
Metropolitan C5 centers National retail chains Multi-site logistics Industrial operations
Works with what you already have

Any standard IP camera. No exceptions worth listing.

If it speaks ONVIF, RTSP, or HTTP MJPEG, Omnisight ingests it. That's the entire commercial CCTV market for the last decade plus, from 480p legacy fixed cams to 8K PTZ. No proprietary firmware required. No vendor lock-in. Most deployments are mixed-fleet by camera-month two.

ONVIF Profile S/T/G
RTSP / RTP
HTTP MJPEG
H.264 / H.265
Tested on the major brands
Axis Hikvision Dahua Bosch Hanwha Avigilon Pelco Uniview i-PRO Verkada Ubiquiti generic ONVIF
89% of CCTV footage
goes unwatched
1 : 100+ operator to camera ratio
at most sites

Most places have more cameras than operators watching them.

Cameras stopped being eyes a long time ago. They became recorders. The footage gets reviewed after something has already gone wrong. Omni watches every camera at the same time, at the level of detail a trained operator would, and tells you the moment something matters. In real time, not in hindsight.

The AI operator

Meet Omni.
Talk to your surveillance.

Stop clicking through 200 cameras. Ask Omni in plain language, and it pulls the answer with evidence attached. Type, voice, or chat.

app.omnisight.ai/omni/demo LIVE

Sample queries: "Find a blue truck near the OXXO on Insurgentes between 6 and 7am yesterday." "Who was the last person to enter the loading dock?" "Notify me when a car parks longer than 15 minutes in the red zone."

What Omni can do

What Omni sees.
Everywhere. All the time.

A single perception engine, running every capability below in parallel on every camera you have. Continuously, without supervision, without missing.

01

Object detection & classification

People, vehicles, weapons, packages, animals, and any custom class your operation needs. Tuned on the actual footage your cameras produce.

02

Face recognition

Identify or alert against opt-in watchlists. Stored as embeddings, not photographs. Privacy controls and policy-based redaction available per deployment.

03

License plate recognition

Reads plates across US, EU, LATAM, and Asian formats. Watchlist matching runs in under 100ms with a full audit trail. Hit rate above 96% on real CCTV feeds.

04

Object dossier extraction

For every detected object: attributes, direction, pace, time on frame. People get clothing and demographics. Vehicles get make/model/color/speed. Each row indexed and queryable.

05

Action detection

Not just objects: what they're doing. Person falls. Vehicle strikes pedestrian. Bag is dropped and abandoned. Two people start fighting. Multi-frame action models trained on real incidents.

06

Cross-camera re-identification

Track the same vehicle or person across hundreds of cameras. The investigator searches by description or last-known sighting. Three-day manual traces collapse to minutes.

07

Spatial relations

Search for objects in relation to other objects or landmarks. "Blue truck near an OXXO." "Anyone within 3 meters of the loading dock." "Vehicle stopped under the overpass."

08

Rules & instructions

Write rules in plain English. "Notify me if you see someone holding a weapon." "Alert when anyone enters the loading dock after midnight." "Flag any vehicle parked longer than 30 minutes in the red zone." Omnisight enforces it across every relevant camera.

09

Natural-language visual search

"White SUV with pine trees on the top of the car." "Person in a yellow hoodie carrying a backpack." Describe what it looks like, in any language, and Omnisight finds it across every camera and every hour of footage.

10

Camera obstruction detection

A covered lens, a defocused image, a paint attack, a deliberate knock. Sharpness loss is detected within seconds, not the days most networks take to notice.

11

Built-in VMS & PTZ control

A unified view of every camera, regardless of vendor. Pan, tilt, zoom, set presets, all via ONVIF. Stop juggling four control panels for four manufacturers.

12

Scene text recognition

Reads every piece of text in the frame: storefront signs, billboards, banners, vehicle wraps, parking signs, posted notices, even handwritten signs. Each read indexed with location and timestamp. Search any scene later by what was written in it.

How it runs

Production specs that matter.

The numbers procurement, security, and operations leaders ask about first. Detailed architecture diagrams available under NDA.

Latency
Median <350ms from frame capture to alert. P99 under 800ms. Measured end-to-end on production deployments, not just model inference.
Accuracy
Vehicle detection above 97% recall on real CCTV footage. ALPR above 96% on plates of acceptable quality. Cross-camera ReID above 89% on tracks of 4+ cameras. Per-deployment accuracy reports issued on request.
Throughput
Single deployment scales to 10,000+ cameras. Per-camera cost falls at scale. Hardware sizing guide and reference architecture available under NDA.
Training data
Models are tuned on real operational footage from active deployments, not public benchmark datasets. Each environment goes through a calibration phase before going live.
Deployment models
Cloud, on-premise, sovereign, or air-gapped. Customer-managed keys available. Footprint configurable from single-site pilots to multi-region production.
Compliance
SOC 2, ISO 27001, FedRAMP-ready. CJIS-aligned (US), GDPR-aligned (EU), LFPDPPP (Mexico). On-premise configurations available for jurisdictions with data residency requirements.
Camera support
Any RTSP-capable camera. No hardware replacement required. Tested on networks ranging from 2015-era IP cams to current 4K PTZ. Integrates with existing VMS / NVR / camera-management infrastructure.
Languages
Operator interface and natural-language search in English and Spanish. Additional languages added on customer request.
Integrations

Works with what you already run.

Omnisight is a software layer, not a camera company. Three honest paths, six concrete entry points.

RTSP streamsIP cameras
ONVIFDiscovery & control
REST APIDocumented · v1
WebhooksEvent-driven
CAD integrationsCustom · dispatch
Custom workflowsBuilt on API
Common questions

What buyers actually ask.

Do we need to replace our cameras?

No. Omnisight runs on top of whatever you already have. If your cameras output RTSP, which virtually all commercial CCTV does, we can ingest the feed. The cameras don't need any AI capability built in. We've deployed on networks with cameras ranging from 2015 IP cams to 2024 4K PTZ.

How is this different from what our VMS already does?

VMS systems record and let you search; Omnisight understands what's happening as it happens. A VMS will tell you "there's footage of the front gate from 2:14am." Omnisight will tell you "a silver Vauxhall Astra entered the front gate at 2:14am, heading east at 24mph, plate AB52 XYZ, not on any watchlist." It's the AI layer on top, not a replacement.

What about privacy and data residency?

Customer-managed keys are available on all deployments. Mexico deployments meet LFPDPPP requirements with in-country processing. EU deployments are GDPR-aligned with data never leaving the region. US public-sector deployments follow CJIS-aligned controls. On-premise and air-gapped configurations are supported for the most sensitive environments.

How accurate is it, really?

On real CCTV footage (not benchmarks) vehicle detection runs above 97% recall, ALPR above 96% on plates of acceptable quality, and cross-camera ReID above 89% on tracks of 4+ cameras. We publish per-deployment accuracy reports because every customer's camera quality and lighting is different. We commit to numbers we can defend in front of a procurement officer.

What does deployment look like?

For most customers: 2 weeks from signed contract to live on the first 20 cameras. The first week is infrastructure provisioning and integration with the existing camera network. The second week is calibration on real customer footage. Every site's "normal" is different, and the system learns it before going live. A pilot can be running in under 5 days if the infrastructure is ready.

What does it cost?

Pricing is per-camera per-month with volume tiers. The more cameras, the lower the per-camera cost. We publish a pricing matrix to qualified prospects under NDA. Most enterprise deployments land between $33 and $69 per camera per month, depending on capability tier and volume. Government deployments use a different structure, typically per-jurisdiction-per-year.

See it on your camera feed.

We'll connect Omnisight to one of your cameras in a sandbox and show you what it detects in 30 minutes, without committing to anything.

V1V2