PLATFORM
The reasoning layer for the first mile.
A container on the router already in the vehicle. It reads what the crew says, what the scene shows and what the equipment reports — and produces one record of what happened, whether or not the network is up.
WHAT GOES IN
Four signal classes, one timeline.
Voice
Ambient cabin audio
radio and intercom
Video
in-cabin and forward
thermal
Device telemetry
patient monitors
vehicle bus, IoT sensors
Domain context
dispatch, GPS
network telemetry
WHAT THE PLATFORM DOES
Four capabilities, one container.
01
Capture at the source
Every stream, on the vehicle’s own network
- Devices connect to the container over the vehicle’s local network
- Documented vendor interfaces where they exist
- Sensors we supply, or the cameras and microphones already on the vehicle
Why it matters. Deployment does not wait on a device vendor to open an interface. Days, not quarters.
02
Deterministic signal fusion
- Time alignment across voice, video, telemetry and location
- Mission state derived from motion, geofence, dwell and vehicle signals — no dispatch integration required
- Streams reassembled into one time-aligned account of the run
Why it matters. Telemetry alone registers that a value changed. It never says what was happening.
Edge reasoning
- Small language models running on the vehicle, sub-100ms local inference
- Cloud used for enrichment and validation, never in the critical path
- Where inference runs is a tenant setting — edge-only where sovereignty requires it
Why it matters. The edge stays intelligent at zero backhaul. Crews keep working through the drop.
Governed output
- Every assertion carries the timestamp and reference of the moment it came from.
- Classification travels with the record; egress policy is enforced in the container
- Structured output typically 8–16 GB per vehicle per month — over 90% less than raw media
Why it matters. An auditable record, and a bandwidth bill that survives scale.
WHERE IT RUNS
A substrate is a host adapter, not a fork.
SUBSTRATE
STATUS
NOTES
Ericsson Cradlepoint
certified
R1900 and R2400. Deployed and upgraded through NetCloud Manager — no new hardware class, no second management plane.
x86 edge compute
in production
Generic hardware with cellular dongles, where no vehicle router is present.
Industrial gateways
host adapter
Cisco IOx, ruggedised gateways, private 5G and MEC nodes.
Fixed network edge
host adapter
Container-capable switches in plants, terminals and command facilities.
The container is the same on every substrate. What changes is a host adapter that reads the platform’s own telemetry — signal quality, GPS, ignition, attached devices — and hands it to the pipeline unchanged.
One record, many renderers.
01
A structured record
Industry-specific output rendered from one canonical representation. The renderer is deterministic and runs with no model calls and no network access.
02
An evidence chain
Every fact traceable to the moment it happened, with version history, model reference and the signature of the person who approved it.
03
Sovereign egress
What leaves, in what form, to which destination, under whose policy — authorised centrally, enforced in the container, and logged as a decision.
04
Operational readiness
Sub-100ms local reasoning, and continuous capture through the network drops that occur on every run.
Adding an industry is a config file. The middle never changes.
Privacy by design — we never store recordings.
Audio and video are processed on the vehicle and discarded. Training uses de-identified records and corrections only, and contributing beyond the customer’s own model is opt-in.