KoiVision® Platform · Yard Intelligence Suite
AI yard intelligence for the gate, the yard and the dock.
A complete digital control tower for the yard. Gate cameras and yard mules keep a slot-level picture of every trailer true, KoiAssist acts on it and KoiGPT explains it — written straight into the YMS, WMS and TMS you already run. Generally available, and live at Fortune 50 scale.
Gartner® Cool Vendor 2026 — Intralogistics Smart Robotics
NVIDIA Partner
NVIDIA showcase, NRF 2026 and NRF 2025
Live at Fortune 50 scale
The short answer
What is yard intelligence?
Yard intelligence is the use of computer vision and AI agents to keep a live, verified record of every truck, trailer and container moving through a facility's gates, yard and dock doors — and to act on that record. KoiVision does it with cameras at the gate and on the yard mules that already drive the yard, writing slot-level location, condition and dwell into your existing YMS, WMS and TMS.
A yard management system (YMS) is the record. Yard intelligence is what keeps that record true without a person walking the yard with a clipboard.
Key facts
- Product
- KoiVision Yard Intelligence Suite, by KoiReader Technologies, Inc. (Dallas, TX)
- Status
- Generally available; live at Fortune 50 scale
- Scope
- Gate, yard and dock — nine products, one platform
- Sensor
- Mule-mounted cameras on NVIDIA Jetson Orin AGX. No RFID grid, no BLE beacons, no trenching
- Gate
- 30-second gate check-in with the AI kiosk
- Integrates
- YMS, WMS, TMS, ERP and SAP EWM
- Recognition
- Gartner® Cool Vendor 2026; NVIDIA Partner
What operations teams ask us to fix
Where the yard costs you money
Dwell and detention
Trailers sit, carriers bill, and nobody can show why.
Gate queues and dock turns
Manual check-in backs trucks up and slows every door.
Trailer inventory visibility
A trailer nobody can find can stop a production line.
OS&D and damage claims
No timestamped condition at entry means disputes drag on.
Labor at the gate and yard
Clipboard and guard processes that don't scale with volume.
A blind YMS / WMS
Systems of record that are only as current as the last walk.
How it works
A yard is an airport. It needs a control tower.
KoiReader's four-layer approach is a complete digital control tower, built from the ground up. Each layer earns the next — and every layer is in production today.
Perception is about to stop being the hard part. The winners won't have the best detector; they'll have the picture of reality that stays true.
Interface
Explain it KoiGPT
Agent
Act on it KoiAssist
Twin
Prove it KoiTwin
Perception
See it Gate cameras + yard mules
The suite
Nine products, one platform
Vision and agentic AI for the gate, yard and dock. It reads physical flow and pushes it into the systems you already run.
Gate
KoiDriver
Mobile app, QR and Smart Tablet gate automation for drivers.
Gate
KoiKiosk
AI kiosk check-in with ERP, YMS and TMS validation.
Gate
KoiGuard
Vision AI at the gate for trailer integrity and CTPAT.
Yard · See it
KoiRover
Mule-mounted, slot-level yard asset tracking.
Yard · Prove it
KoiTwin
LiDAR digital twin of the yard for provable optimization.
Yard · Act on it
KoiAssist
AI agent that directs yard-mule moves and scans.
Explain it
KoiGPT
Domain LLM for yard-efficiency questions and answers.
Integration
KoiConnect
TMS, YMS and ERP integration across the full lifecycle.
Gate · Upcoming
KoiPass
CLEAR-style touchless gate.
See it · Perception
How does trailer tracking work without RFID? The yard mule is the sensor.
The yard truck already drives everywhere, so we made it the sensor. PoE cameras on the mule, with NVIDIA Jetson Orin AGX on board, read every slot on every pass and confirm each trailer with the frame on record.
At the gate, cameras and our kiosk check in every visit, with condition photographed in and out. One decision removes an entire infrastructure project from the capital plan.
✕No RFID grid
✕No BLE beacons
✕No trenching
One layer down — a single NVIDIA DeepStream pipeline
NVDEC
Decode on the GPU
NVMM
Device-resident buffers
nvstreammux
Every stream, one call
TensorRT
Detect every frame
Tracker
Persistent-ID tracking
Quality gate
Drop the redundant
Triton
Recognition head
One big model over every frame wouldn't fit a sealed box on a moving truck. The cascade is why this runs at the edge at all — and right down to the individual slot.
Mule-mounted vision vs. RFID, BLE and drones
| KoiVision (mule is the sensor) | Traditional RFID / BLE / drones | |
|---|---|---|
| Infrastructure | Cameras on equipment that already moves. No grid, beacons or trenching | Tag, reader or beacon grid to install and maintain |
| Freshness | Near-real-time yard truth, every pass | Dashboards up to eight hours stale |
| Read quality | Slot confirmed, with the frame on record | 25–35% shrinkage and false reads |
| Scale | CPG, grocery, beverage, auto/VPC, paper, port | Does not scale like mule-mounted vision |
Prove it · KoiTwin
A yard digital twin you can test policy in before you change the yard
You can't A/B test a live yard. KoiTwin starts from a LiDAR base layer with slot-level precision — about 7M sq ft captured by drone in one day — and makes optimization provable before deployment.
1 · Observe
Gate cameras and mules keep the twin true, every shift.
2 · Simulate
Try a door policy, mule count or cuOpt plan on real history.
3 · Ship
Roll out only what's been proven.
4 · Measure
Compare the live result with the simulation.
Built on NVIDIA Omniverse and OpenUSD. NVIDIA Cosmos generates synthetic snow, night, damage and rare events, so cameras learn them before the yard sees them.
Act on it · Explain it
KoiAssist closes loops. KoiGPT opens them up for inspection.
Same perception underneath both — every answer and every move traces back to a frame.
KoiAssist — the AI yard agent
Everyone has seen computer-vision demos. KoiAssist takes a confirmed record, dispatches a physical move to the driver and writes it back to the system of record. The mules see the result on their next pass — the loop closes without anyone chasing it.
- Sense: yard state by slot and dwell, gate events, dock door status from YMS
- Decide: your operating rules plus NVIDIA Nemotron Super reasoning
- Act: move a trailer to a door or slot, rebook appointments, check in and out
- Record: live slot info, seal and damage checks into YMS, WMS, TMS and SAP EWM
Next: NVIDIA cuOpt for slot at check-in, dock door scheduling, move assignment and delivery-window sequencing.
KoiGPT — ask your yard anything
A domain LLM above the operational loop, not in the control path. It reads everything and writes nothing, so a question can never change the yard.
It is grounded in both your data (yard, gate, appointments) and your documents (SOPs, carrier agreements, rules). Data alone gives generic advice; documents alone give theory. The value is in the intersection.
Retrieve · Join · Ground · Compute · Explain
Example: a gate manager at a major automotive plant, the morning after a holiday. No dates, no context.
Why was check-in slow yesterday?
Yesterday was the first day back after Labor Day. Gate 20 ran at twice its normal check-in time, 06:00–09:00: the holiday backlog landed in three hours, one truck in four had no appointment and needed manual approval, and Gate 6 opened overflow only at 09:30.
Recommendation: open Gate 6 overflow at shift start after any holiday, per your gate SOP.
Evidence: 200 check-ins · holiday calendar · appointment log · gate SOP
In production
One live deployment, one month
At a major automotive assembly plant.
~30,000
gate transactions in September
~1,400
gate transactions every weekday
~1 min
arrival to fully approved, sign-offs included
Read the Fortune 50 yard intelligence story →
What it's worth
What is yard intelligence worth?
Illustrative models, not customer results — sized for your site in the working session.
In manufacturing
A large automotive assembly plant with four manned gates.
Gate labor, four gates
~$0.7–1.15M / yr
Inspection and claims
~$190K / yr
One avoided 8-hour line stop
~$10M
One avoided 8-hour stop is worth roughly a decade of gate labor savings.
In retail and CPG
Protect OTIF. Then cut gate cost.
~$0.7M–$8M+ / yr
OTIF and network-compliance exposure: a yard choke causes late deliveries, which draw a 3% COGS charge on assessed goods.
- Gate labor and detention
- Yard inventory and dwell
- Line-down and supplier visibility
- Carrier billing from gate scans, not late dock check-ins
Illustrative model: market wage build-up ($17/hr, 1.7× burden), industry claim and downtime benchmarks, and public 10-K and diligence estimates. Ranges cover 24/5 and 24/7 gate schedules. Not named KoiReader customer results.
Accelerated by NVIDIA
GPU-accelerated from the mule to the cloud
Physical perception on the yard mule and at the fixed cameras; agents and LLMs at the edge or in the cloud. Same software at every tier.
Yard mule
NVIDIA Jetson Orin AGX
Jetson embedded. Thor in active evaluation.
Fixed cameras
L4 · A6000 · RTX 4500 Pro
Edge, at the gate and on site. Sized to the streams.
Edge / cloud
H100
Agent and LLM components, plus training.
Next
Blackwell
Tested against our suite ahead of deployment.
DeepStream
TensorRT
Triton
Omniverse · OpenUSD
Cosmos
Nemotron Super
TensorRT-LLM
Dynamo
NeMo Guardrails
cuOpt (next)
One platform, already live on
Gate · Yard · Dock · Warehouse · Factory floor
Works with the systems you already run
KoiConnect writes into existing YMS, WMS, TMS and ERP — including SAP EWM.
Proof
- NVIDIA partner, built on NVIDIA Metropolis (NVIDIA blog on KoiVision)
- Six granted U.S. patents
FAQ
Yard intelligence: common questions
Does KoiVision replace my yard management system (YMS)?
No. KoiVision reads the physical flow at the gate, yard and dock and pushes it into your existing YMS, WMS and TMS. Your system of record stays in place and becomes accurate on its own.
Do I need RFID tags, BLE beacons or new infrastructure?
No. Cameras on the yard mule read every slot on every pass, and cameras at the gate check in each visit, so there are no RFID tags, BLE beacons, GPS trackers, drones or trenching to buy, install or maintain: No RFID, No BLE, No Drones.
How accurate is trailer location tracking?
Mule-mounted cameras read every slot on every pass and confirm each trailer with the frame on record, so every location comes with photographic evidence. Slot-level results are sized for your site in the working session.
How fast is gate check-in?
KoiReader's gate automation and AI kiosk are built for a 30-second gate check-in. Cameras read tractor and trailer details on approach, the kiosk validates the driver and paperwork against your systems, and condition is photographed in and out.
What is the difference between KoiAssist and KoiGPT?
KoiAssist is the agent: it takes a confirmed record, dispatches a physical move to the driver and writes the result back to your system of record. KoiGPT is the interface: a domain LLM that reads everything and writes nothing, so a question can never change the yard. Both rest on the same perception, so every move and every answer traces back to a frame.
Which industries use yard intelligence?
Manufacturing, retail and CPG, grocery, beverage, auto/VPC, paper and port operations. One live deployment today is at a major automotive assembly plant.
What hardware does it run on?
Yard mules carry PoE cameras with an NVIDIA Jetson Orin AGX on board. Fixed cameras at the gate and on site run on NVIDIA L4, A6000 or RTX 4500 Pro GPUs, and agent and LLM components run on H100 at the edge or in the cloud. Same software at every tier.
How do we get started?
Book a 20-minute working session. Bring your gate, yard and dock, and we'll show you where the yard is costing you and what KoiVision would change first.
Who uses KoiReader?
KoiReader is in production at Fortune 50 enterprises, is an NVIDIA partner building on NVIDIA Metropolis, and holds six granted U.S. patents, including ones for yard mapping and asset tracking and for monitoring warehouse or yard inventory.
Twenty minutes. One facility. Your numbers.
Bring your gate, yard and dock. We'll show you where the yard is costing you and what KoiVision would change first.
Explain it
Act on it
Prove it
See it
The complete digital control tower for the yard
