The Fogsphere platform
Product
Fogsphere is an end-to-end AIoT platform. Edge+ runs inference on site, Manager orchestrates the fleet and holds the evidence record, and MistIQ trains and ships the models that keep both current.
One platform, from camera to boardroom
Video slot — awaiting upload.
Architecture
Three tiers, one platform
A hyper-scalable Kubernetes platform for distributed deployments across edge nodes, on-premises data centres and cloud regions. Edge-to-centre traffic is event-based, so the link between site and Manager stays light enough for 5G and satellite backhaul.
Tier 1 — on site
Fogsphere Edge+®
- Ingests video and sensor data from site cameras and LoRaWAN devices
- Executes AI models and processes LoRaWAN sensor data in real time
- Delivers AI and IoT events to Fogsphere Manager
- Executes automations that send data and commands to OT systems
- Sends SMS or email notifications through the customer’s own systems
- Deploys on a gateway or server with one or more GPUs, sized by camera and model count
Tier 2 — site, data centre or cloud
Fogsphere Manager®
- Stores AI models and event data, with retention rules per zone and per case
- Configures Flows® and deploys them to Fogsphere Edge+ nodes
- Provides visual analytics and Vision Agent® compliance reports
- Manages every AI and IoT asset on a single global map
- Includes tooling to build analytics dashboards without writing code
- Interacts with IT systems, including single sign-on
- Scales with Kubernetes; needs GPUs where Vision Agent® is enabled
Tier 3 — training
Fogsphere MistIQ®
- Infrastructure to build, train and benchmark AI models
- Trains and delivers VLM and CNN model updates over the air
- VLM/CNN data annotation on your own footage
- Datasets management and versioning
- Distributed training workers, so jobs run without moving your data
- Deploys on internal or hyperscale cloud infrastructure
Fogsphere Manager®
What sits inside Manager
Manager is where models, events, assets and dashboards live. Six components do the work, and each one is usable by an operations team without a developer in the room.
Low-code AI Flow® Editor
Build a detection Flow per camera or per zone by dragging blocks, not by writing code. Flows are versioned centrally and pushed to Edge+ nodes.
Insights Studio®
A visual dashboard editor. Compose the charts, tables and live tiles a site manager or an HSE lead actually needs, and publish them per role.
Asset Manager — integrated Digital Twin
Unifies cameras, Edge+ nodes, sensors and open cases on a single global map, exposed over APIs and able to feed enterprise digital twins.
AI Case Manager & Forensics
Groups related detections into a case, holds the clips and metadata, and lets an investigator search back across historical footage by attribute.
Vision Agent® with RAG on VLMs
An agentic layer that reasons over the event record using vision language models, so operators can ask questions of the estate in plain language.
LoRaSync® LoRa Network Server
A built-in LoRaWAN network server. Gas, temperature, water and vibration readings land in the same event stream as the camera detections.
Fogsphere Edge+®
The edge node that does the inference
Edge+ is the software stack that turns a GPU server or gateway on your site into a fleet-managed inference node. Video never has to leave the site for a detection to be raised.
Edge AI Software Stack Management
Model images, runtime and Flows are deployed and rolled back centrally. A node that drops off the network keeps inferring and reconciles when the link returns.
Network Management
Camera discovery, stream health, bandwidth and interface configuration handled from the node’s own interface rather than by hand on the box.
Local Events Dashboard
A site-local view of live detections, so a control room keeps visibility even when the link back to Manager is down.
Web Management Interface
Browser-based administration of the node — streams, models, VRAM headroom and automation rules — with no separate client to install.
Secured by TPM®
Hardware root of trust on the node. Stack integrity is measured at boot and model artefacts are bound to the device.
Fogsphere MistIQ®
Training, annotation and benchmarking
MistIQ is the training tier. It exists so that models can be tuned on your own footage, measured against your own scenes, and shipped to the fleet over the air.
VLM/CNN Annotation
Label footage for both convolutional detectors and vision language models in one tool, with the dataset staying inside your infrastructure.
Datasets Management
Version, split and track datasets so a model can be traced back to exactly the frames it learned from.
Distributed Training Workers
Training jobs are scheduled across a pool of GPU workers. Data stays where it is; the job travels to it.
AI Models Benchmarking
Compare candidate models on your scenes before they reach a site, and record the score that justified the rollout.
Hardware and integrations
What it runs on, and what it talks to
Fogsphere runs on standard industrial hardware from Dell, HPE, Lenovo and Supermicro on NVIDIA GPUs, and is delivered turnkey through a vetted network of regional and global systems integrators.
CCTV cameras
Leverages the CCTV infrastructure already on site; partners add cameras only where a position is missing. Any ONVIF-compatible camera, minimum 2 MP, 4 MP or higher suggested — Avigilon, Axis, Bosch, Dahua, Panasonic, Pelco, Uniview and Vicon among them.
NVIDIA GPUs
One GPU with 32 GB+ VRAM for CNN workloads and one for VLM workloads, on DGX Spark and Jetson Thor class devices at the small end and rack workstations at the large end.
LoRaWAN sensors
Off-the-shelf Class A, B and C sensors from all the major vendors, terminated by the built-in LoRaSync® network server.
PLC, DCS and SCADA
Reads state from and actuates control systems over ModBus-TCP and OPC-UA, so a detection can drive an interlock or a signal rather than only an alert.
VMS
Works alongside Milestone XProtect, Genetec Security Center and Avigilon Control Center. Direct camera connection is always advised where it is available.
ERP and HSMS
Certified connectors for SAP, Oracle ERP Cloud and NetSuite, Epicor, Sage, Acumatica, Plex and Odoo; and for Intelex, Sphera, Cority, VelocityEHS and DNV Synergy Life. Custom integration via API through the partner network.
Deployment and sizing
Sizing the platform to the estate
Sizing follows stream count, not site count. The reference designs below are validated starting points; a Fogsphere Global Partner produces the site-specific bill of materials.
Edge+ node — site level
From 10–25 video streams on two or three DGX Spark or Jetson Thor class devices, up to 200–250 streams on a dual-socket workstation with four RTX Pro 6000 96 GB GPUs. Add nodes to handle larger stream volumes.
Manager — central data centre or cloud
A single server carries 10–25 cameras on 16 cores, 128 GB RAM and one GPU, scaling to 200–250 cameras on 48 cores, 512 GB RAM and seven 1.92 TB SSDs in RAID 5. Ubuntu 24.04 throughout.
Clustering and availability
Between 250 and 1,000 event-generating cameras, a minimum three-node cluster delivers 99% to 99.9% availability. Beyond 1,000 cameras the node count grows with the estate.
MistIQ — training tier
VLM training is sized at 48 cores, 512 GB RAM and four RTX Pro 6000 96 GB GPUs; CNN training at 32 cores and 256 GB RAM on the same GPU complement. Storage is dataset-dependent.
One Flow per camera
Sizing assumes one Fogsphere Flow per camera, or several concurrent models bounded by the device’s available VRAM. VRAM headroom is visible in the Edge+ monitoring panel.
Detailed sizing on request
The full per-tier specification tables, certified NVIDIA edge device list and integration matrix are available on request. Contact us for a site-specific BOM.
Platform capability
Evidence, audit and reporting
Evidence capture, retention, reporting and dashboards are platform capabilities in Fogsphere Manager. They span every application rather than sitting inside one, which is why every cluster page points here.
Timestamped evidence
Every detection is logged with a clip, a still, the zone it happened in and the shift it belongs to. The record is written at the moment of detection, not reconstructed afterwards.
Case files and forensics
AI Case Manager groups related detections into a case and holds the supporting media. Investigators search historical footage by attribute rather than by scrubbing timelines.
Retention and privacy
Retention windows are set per zone and per case type. Inference runs at the edge, so raw video stays on site unless a clip is explicitly retained as evidence.
Scheduled reporting
Compliance reports are generated on a schedule and broken out by contractor, zone, shift or asset, so the same numbers reach HSE, operations and the client.
Dashboards per role
Insights Studio® publishes a different view to a control room, a site manager and a group HSE lead from the same underlying event stream.
Audit and dispute support
The event record gives operators visibility for claims, disputes, insurance reviews, regulator audits and client reporting, with the underlying media attached.
Where it runs
One platform, two ways in
The same three tiers sit behind every application and every industry. Start from the capability you need, or from the estate you operate.
By capability
Applications
Thirteen application clusters, from PPE and personal compliance through perimeter security to process and quality assurance — 59 turnkey detections in total.
By sector
Industries served
Nine industry hubs — oil and gas, construction, power generation, manufacturing, healthcare, chemical, logistics, retail and smart city — each with its own compliance lens.
Ready to see the platform on your own cameras?
We will size an Edge+ node against your stream count and run a proof of value on the cameras you already have.