Conventional compression trades quality for space. Ours is built so that the video management platform, the operators, the auditors and the AI models all behave exactly as they did before — while the storage bill falls by up to 90%.
Every large video network we are asked to fix is failing on at least one of these three axes. They are usually described as separate projects; they are the same project.
Statutory and policy retention periods are extended faster than storage budgets are approved. Every extension multiplies an estate that is already at its physical limit.
Adding arrays means adding racks, power and cooling. Remote sites frequently have no room and no viable network path at all.
The features analytics need are precisely the high-frequency detail that aggressive compression discards first.
These are physical properties, not marketing positions. The four invariants protect the indexing, playback and retrieval logic of the incumbent platform. The two lossless guarantees protect the people and the machines that have to use the footage.
No upscaling, no downscaling. Pixels map one-to-one to the original.
No frame dropping. The original temporal smoothness is preserved exactly.
H.264 and H.265 are carried through natively, so no client needs a new decoder.
Millisecond-level alignment keeps timeline scrubbing and seek positioning exact.
No perceptible degradation for operators, investigators or reviewers — validated by no-reference quality scoring at frame level.
Person, vehicle and plate features are retained so downstream structured analysis is entirely unaffected.
Why both matter. An operator reviewing an incident and a model detecting a plate are two different consumers of the same pixels. Systems that only satisfy one of them create a silent failure: footage that looks acceptable and cannot be used.
We do not amplify grain or invent detail. We remove information the eye and the model both discard — then rebuild the rest faithfully.
Human sensitivity to luminance and chrominance differs. Dynamic adjustment of colour and brightness removes what the eye cannot resolve.
Adjacent pixels in an image are highly similar. Removing that repetition reduces data volume without touching structure.
Consecutive frames share large amounts of content. Inter-frame coding removes the duplication that carries no new information.
Key ROI preservation. On top of the three redundancy classes, AI-assisted encoding detects regions of interest — faces, vehicles, plates — and allocates bitrate preferentially so that the features downstream analytics depend on survive the process at full fidelity.
The hardest case is not a new build — it is an existing network with millions of files already on storage and a management platform that must not be disturbed. X-Stream Archive was designed around that constraint.
A clear boundary lowers systemic risk and physically protects the evidentiary value of the footage. These exclusions are architectural, not configurable.
Each file passes through an isolated working area. If any gate fails, the temporary file is destroyed automatically and the source file is left completely untouched.
The temporary file's container and basic playback behaviour are verified before the output is trusted for anything else.
Machine-level comparison of resolution, frame rate, codec and duration between the new and original files. All four must match exactly.
PSNR and SSIM are evaluated against the expected envelope. Falling short triggers a preventive circuit-breaker instead of a replacement.
Preventive circuit-breaker. If picture degradation is detected — PSNR or SSIM below the expected envelope — processing halts before the source file is touched. Repeated failures permanently stop processing that file, safely retain the original, and raise a system alarm, avoiding the I/O risk of repeated attempts.
Hardware independence is not an optimisation — it is what makes the system deployable in a sealed government network and affordable at city scale.
The platform analyses day and night, lighting, motion level and scene complexity, then applies the strategy that scene actually needs. No operator parameter tuning.
City roads, checkpoints and commercial streets. Strategy leans toward stable quality, protecting the detail of people and vehicles.
Corridors, warehouses and fixed entrances. Strategy leans toward a higher compression ratio to release the maximum storage.
Smart campuses, car parks and public service halls. Strategy balances space against quality for footage that changes character through the day.
Most clients start with one and expand into the others as coverage grows. There is no re-engineering step between them.
Compress at each distributed site before the stream leaves the premises. Removes the need for local NVR stacks and slashes backhaul bandwidth.
A mid-size campus or industrial site served by a single appliance carrying compression, AI inference and a lightweight VMS in one box.
City, province and carrier scale, converging hundreds of streams per node with N+M hot standby and service-level failover.
Zero new hardware. Compress in the platform on general-purpose CPU, with linear horizontal expansion as ingest grows.
Compression is not only a storage saving. It reduces the hardware, rack, power, bandwidth and maintenance lines together — and it preserves AI capability rather than trading it away.
Per 1080P stream, 30-day retention
Storage, rack, power and bandwidth footprint
Per 1080P stream, 180-day retention
Storage, rack, power and bandwidth footprint
Tell us your camera count, resolution mix, retention requirement and network topology. We will return a target architecture and an indicative capacity model.
