Home Products AI BOX
Edge AI BOX · On-site Compression + Face Indexing

AI BOX

The moment you move to 4K, retention days collapse first.

Retention days = storage capacity ÷ daily footage volume. Raising resolution multiplies the daily volume, and on the same storage your retention days shrink by the same factor. AI BOX compresses the video right next to the camera, shrinking that denominator. Your cameras stay. Your existing VMS stays.

Request a Demo Request a Retention Calculation

The real problem isn't capacity — it's retention days

What operators fear is not the storage invoice. It is the moment someone asks for footage from that day and the only answer is: it's gone.

A horizontal bar chart showing retention days falling from 30 days to 7.5 days as image quality is raised on the same storage Same storage · the higher the image quality, the fewer retention days Operating policy line: 30 days FHD 4Mbps 30 days · meets policy 4K 8Mbps 15 days 4K 12Mbps 10 days 4K 16Mbps 7.5 days Retention days when storage is unchanged and only image quality is raised

Thirty days is not a legal figure — it is the usual operating policy of public bodies and companies. When an incident, complaint, or investigation request arrives, whether that day's footage still exists stops being an equipment issue and becomes the operator's accountability issue.

A horizontal bar chart showing retention days at 4K 16Mbps: 7.5 days if left as is, 15 days with a codec upgrade, and 50 days with AI BOX At 4K 16Mbps · what gets you back over the 30-day line Operating policy line: 30 days If left as is 7.5 days Codec upgrade (about −50%) 15 days · still short AI BOX (about −85%) 50 days · recovered Same storage · same 4K 16Mbps quality

Adding storage grows the numerator; compression shrinks the denominator. Once you fix the goal at "30 days at 4K," the two approaches produce very different bills.

"A codec upgrade buys you roughly double, and stops there. Double is not enough to cross the 30-day line."
The differentiator is not how much you reduce — it is whether you clear the line or not.

Not a rip-and-replace project — a drop-in

Exactly one new block goes in, between the cameras and the storage. Everything on either side stays as it is.

A three-stage flow in which only one new block, AI BOX, is inserted between the existing cameras and the existing storage Unchanged Existing cameras Any brand · no replacement The one new block AI BOX On-site compression + face index 4 channels per box Unchanged Storage · existing VMS Kept in place, runs alongside ONVIF · RTSP integration / no central GPU server

Because compression finishes right beside the camera, no new central server room is required. Sites without a server room are in fact our typical customers.

01
Keep Your Cameras

Connects over ONVIF and RTSP. Camera brand does not matter, and there is no replacement cost.

02
Runs Alongside Your VMS

Your existing VMS is not removed. It sits in between, so day-to-day operation does not change.

03
No Central GPU Server

Compression finishes at the camera. No new server room, cooling, or extra infrastructure.

04
Works When the Link Drops

Processing is local, so compression and recording continue. The network is used only for remote delivery.

Configuration and specifications

Two product lines, chosen according to site conditions.

Line 01
Standard

Built on a low-power edge processor. Handles 4 channels at a sampling rate of 2 frames per second. The standard configuration for control-room sites.

Line 02
High Performance

Built on a higher-tier processor with an NPU. Covers real-time processing and 4 to 8 channels. For sites with heavier processing demands.

Which line fits is confirmed after we review your channel count, processing mode (after-the-fact search versus real time), and installation environment.

ItemSpecification
Camera Connections4 channels per box
ScalingOne additional box per 4 channels (no server expansion)
IntegrationONVIF · RTSP (runs alongside existing VMS)
Supported Formats4K (UHD) / 1080p / H.264 / H.265 / MJPEG
CompressionAbout 85% reduction (measured on site; varies with scene composition)
Measured 4K BitrateAbout 2.1Mbps (about 22.75GB per channel per day)
ProcessorHigh-performance edge processor (NPU)
GraphicsIntegrated graphics acceleration
Memory32GB DDR5
StorageM.2 NVMe 512GB
EncryptionAES256
Dimensions240 × 190 × 50mm
Weight1.9kg (net)
MaterialAluminium alloy case
PowerDC 12V (optional 9V–36V)
Operating Temp-10°C to 60°C
Storage Temp-20°C to 70°C
Humidity10% – 90% (non-condensing)
On how we describe image quality
The accurate phrasing is a level at which quality degradation is hard to perceive, and that judgement should be made by viewing the original and the compressed stream side by side in a demo rather than on a claim.

Where — and by how much — do you actually win

The value is not the same everywhere. We state plainly where the win is large and where you get only one benefit.

4 channels
Cameras handled by one box
~85%
Footage volume reduction, measured on site
2.1Mbps
Measured bitrate per 4K channel
50 days
Retention recovered at 4K 16Mbps
Large win
Wireless-link sites (construction, traffic, rural, wildfire watch, mobile)

Sending video costs money every month. Cutting the transmitted volume cuts that bill directly — and you secure retention days at the same time.

One benefit only
Wired private networks and small sites (control centres, buildings)

Link costs are effectively zero here, so reducing transmission saves no money. What we deliver at these sites is not cost savings — it is retention days, and only that.

Do not compare one unit against one unit. An honest comparison fixes the goal first — "to hold 30 days at 4K, how many storage units must be added versus how many boxes?" That answer requires four site conditions: channel count, storage capacity, retention policy in days, and image-quality standard.

It builds an index while it compresses

Retention that cannot be searched is retention you cannot use.

Holding 30 days is meaningless if a person has to scrub through all 30 days by eye. Retention succeeded; retrieval failed. AI BOX builds a face index while it compresses, so a later search queries the index instead of rewinding the footage. The search finishes in seconds.

Stated up front
  • We narrow the candidates; the final determination is made by your officer. This is not a feature that identifies a person automatically.
  • Surveillance cameras are distant, angled downward, and mixed with night and backlit conditions. Figures measured under optimal conditions do not reproduce as-is in the field.
  • Compression has measured data from public-sector sites; face indexing is still being validated in pilots.
Municipal control centres — where this configuration actually goes in →

See the package that ties retention recovery and after-the-fact search into a single line.

Frequently asked questions

The eight questions we receive most often during evaluation.

How many cameras can be connected?
One box handles 4 channels. For 40 cameras, 10 boxes; for 100 cameras, 25 boxes. As cameras grow, you add boxes rather than expanding servers.
Does it really work without a server?
Compression finishes inside the box next to the camera, so no central GPU server is required. Sites without a server room are in fact our typical customers.
Do we have to replace our cameras and VMS?
No. It connects over ONVIF and RTSP, so camera brand does not matter, and your existing VMS keeps running alongside rather than being removed. Older environments need an integration check, so send us the software name and version, plus how the cameras are connected. We will check and reply.
What happens if the internet goes down?
Processing is local at the edge, so compression and recording continue even when the link is down. The network is used only for remote delivery.
What is the compression ratio? Does it vary by site?
We quote about 85% reduction based on on-site measurement. The more motion in the frame, the less room there is to reduce, so it varies by zone. Tell us foot traffic, vehicle share, and night-time share, and we will prepare a demo matched to those conditions.
Doesn't image quality drop?
Degradation is at a level that is hard to perceive. But this is not something a claim can settle — it should be judged by putting the original and the compressed stream side by side in a demo, using the scene composition you actually operate with.
Couldn't we just add disks or change the codec?
Adding storage grows the numerator; compression shrinks the denominator. If the goal is fixed at holding 30 days at 4K, storage may have to be expanded by several units. A codec upgrade cuts roughly half, but at 4K 16Mbps half still lands around 15 days — short of the 30-day policy.
What does it cost to deploy?
We reply with a quotation. Four inputs are needed to price it accurately: channel count, storage capacity, retention policy in days, and your current image quality plus any 4K transition plan. Send those and we will reply together with a retention calculation.

Related products

Products commonly evaluated alongside this one.

VCA — The video search brain →

Condition-combination search on face, body, colour and carried items, plus companions and routes
GATE — Server-installed biometric authentication →

The terminal stays yours; only the authentication engine goes inside your server
GATE API — The face recognition engine →

1:N authentication, liveness and face indexing, delivered as an API

Move to 4K and still keep your 30 days

Send four things — channel count, storage capacity, retention policy in days, and image-quality standard — and we will calculate how many days your current setup reaches and reply with the result.

Send Conditions, Get the Calculation