Deployment & Architecture

VIDIZMO supports nine deployment models to fit any infrastructure and compliance requirement.

• SaaS Shared -- multi-tenant
• SaaS Dedicated -- single-tenant
• SaaS FedRAMP -- High via ProjectHost
• On-premises
• Private cloud
• Hybrid
• BYOC -- Bring Your Own Cloud
• Air-gapped -- fully disconnected
• Containerized -- Docker/Kubernetes

VIDIZMO supports multiple cloud platforms with Microsoft Azure as the primary cloud.

• Microsoft Azure -- primary cloud with PaaS services
• Azure Government Cloud -- CJIS/FedRAMP workloads
• AWS and AWS GovCloud -- full support
• Google Cloud Platform -- infrastructure-level deployments
• Oracle Cloud -- infrastructure-level deployments

Video Asset Management (VAM) is a system designed to help businesses organize, store, retrieve, and distribute video content efficiently. It acts as a digital library, offering features like centralized video storage, metadata tagging, and secure access control to streamline video management processes.

There is no fixed number. It depends on frame sampling rate, resolution, model size, and precision, and the same GPU can carry roughly 30 cameras or well over 100 depending on those settings. The only trustworthy answer comes from measuring your own model on your own hardware.

On-premises is the default for most fleet-scale deployments, mainly because continuously streaming hundreds of camera feeds to the cloud is a bandwidth problem, and many industries such as CJIS, healthcare, and government require the video to stay on-site. Cloud works well for modest camera counts without heavy sovereignty constraints.

Shared SaaS provisions in approximately 30 seconds via self-service. Dedicated deployments take approximately 4 hours. On-premises and private cloud vary by complexity, typically days to weeks depending on infrastructure readiness.

A VAM system enhances productivity by centralizing video assets, reducing the time spent searching for content. With features like searchability through metadata tagging and easy access for team collaboration, employees can focus more on their tasks rather than wasting time locating videos.

Frame sampling rate is the largest lever by far. Analyzing every fifth frame instead of every frame cuts the GPU's workload to a fifth, and most real detection, like a person crossing a line or a vehicle entering a zone, survives sparse sampling without losing accuracy.

Processing video in the cloud means shipping every camera's stream out of the building continuously, which for a few hundred 1080p cameras is a gigabit or more of sustained, metered outbound traffic. Keeping the analytics on-premises means only small detections and alerts travel off-site, not the heavy video itself.

FedRAMP High compliance through hosting on ProjectHost's FedRAMP-authorized environment. Actively used for VA and other federal deployments requiring the highest security tier for federal cloud operations.

Video Asset Management is crucial for businesses because it helps organize large volumes of video content, ensuring easy access and security. With a VAM system, teams can quickly find videos, improve collaboration, and safeguard sensitive content, making workflows more efficient and boosting overall productivity.

No. Capacity is set by the GPU's compute power and its video decode and encode engines, not by memory. More VRAM lets you hold a larger model or more streams in flight, but past the point where memory is sufficient, extra VRAM does not increase throughput.

Cloud makes sense for a modest camera count where egress isn't a problem, when you'd rather not operate your own hardware, when you're already cloud-native, or when you need to scale capacity up and down rather than size for a fixed fleet.

A fully disconnected installation with no internet connectivity designed for defense, intelligence, and classified networks. All AI processing runs locally on-premises using self-hosted LLMs, and no data leaves the environment.

Metadata tagging in a VAM system allows videos to be categorized by relevant keywords, descriptions, and tags. This makes it easy to search for and filter video content based on criteria like campaign name, project, or topic, improving content retrieval and management.

Yes, quantizing to int8 roughly doubles throughput and halves memory use, at the cost of a small accuracy drop that careful calibration usually recovers. For most detection tasks it functions as close to free additional capacity.

Hybrid keeps the video and inference on-premises, where bandwidth and sovereignty require it, while using the cloud for lighter layers like management dashboards, long-term event search, or specialized models. The heavy video never crosses the boundary.

Docker and Kubernetes are supported for containerized deployments, enabling consistent deployment across cloud and on-premises environments. This simplifies scaling and supports modern DevOps workflows with Infrastructure-as-Code tools.

Any organization that deals with large amounts of video content can benefit from a VAM system. Industries such as marketing, healthcare, education, and corporate training find VAM particularly useful for organizing training videos, marketing materials, and confidential video files.

Because a single streams-per-GPU number silently bakes in one specific combination of model, resolution, precision, and frame sampling. The only way to get a trustworthy number is to measure your own model on the specific GPU with the settings your deployment will actually use.