The central hub between image requests, AI evaluation, your PLCs and your systems.
NexusVisionOrchestrator is an on-premise AI orchestration platform for industrial vision inspection. It routes image-evaluation requests to any AI engine — Maddox Edge, Estek, CodingNexusVision, ONNX, your own, or more on request — and integrates natively with Siemens, Beckhoff and CODESYS PLCs over OPC UA, ADS and S7.
One platform, every line, any AI vendor. Receives image requests from cameras, applications or PLC triggers; routes them to the best engine per request; delivers verdicts to databases, MES, ERP, or back to the PLC.
No sales pitch · Engineer to engineer · 30 minIndependence, protection, scalability.
Strategic value beyond technical features — the long-term reasons platform decisions stick.
Vendor independence
No lock-in. Choose the best engine per task, or run several side by side. Negotiate from a position of choice.
Investment protection
Existing AI tools and CV scripts integrate as adapters. Tomorrow's engines plug in the same way.
Lower integration effort
One platform handles ingest, routing, evaluation, PLC handshake and delivery. Replace N×N integrations with one hub.
Operational scalability
Add lines, cameras, models, engines — without re-architecting. Workers and profiles scale independently.
Future-proof architecture
The AI landscape changes fast. The orchestration layer stays stable while engines underneath are swapped, upgraded or compared.
Lower total cost of ownership
One platform team replaces many siloed integration teams. Pre-gates cut per-inference license cost.
Talk to the line. Directly. No middleware.
Production lines do not run on REST. They run on PLCs. NexusVisionOrchestrator speaks the protocols your controllers already use — OPC UA for vendor-neutral access, ADS for native Beckhoff performance, Sharp7 for direct Siemens S7 access where OPC UA is not licensed.
OPC UA requires the runtime license on the CPU. Sharp7 only needs PUT/GET communication from remote partner enabled — useful for older PLCs without a UA license.
ADS gives native value-change notifications — no polling. Typical jitter ≤ 5 ms. Best fit for tight inspection cycles.
Built-in OPC UA server, no extra license for development. Works with any CODESYS-based PLC including Wago, Beckhoff and Schneider variants.
| Vendor | OPC UA | ADS | Sharp7 |
|---|---|---|---|
| Siemens S7-1500 / 1200 | ✓ recommended | — | ✓ fallback |
| Beckhoff TwinCAT 3 | ✓ via TF6100 | ✓ recommended | — |
| CODESYS V3 runtimes | ✓ built-in | — | — |
Three drop-in flow nodes
plc:read-tag reads a logical tag (Bool, Int, Real, Text) and emits the value as a flow input.
plc:write-tag writes a value back to the PLC — frame acknowledge, status bits, counters.
plc:verdict atomic verdict writeback: Ok / Confidence / TopLabel / Done bits in one transaction.
Soft real-time only. Hard real-time and safety-PLC integration are explicit non-goals.
A live cockpit, not just an editor.
Nodes show live values directly in the body — the last PLC tag, an AI verdict with confidence, a 32-sample sparkline, an inline thumbnail. Port LEDs pulse when data flows. SignalR push at 200 ms. Dark by default, light on a toggle.
- Live by designEvery node shows last value, timestamp and a sparkline. SignalR push, 200 ms cadence — no manual refresh.
- Status at a glancePort LEDs (idle / pulsing / error), category accent stripes, connection-health dots. Find the broken node in seconds.
- Thumbnails inlineImage stages capture an 80×60 JPEG to the snapshot store. See the actual frame, not just a verdict.
- Dark or light, your callIndustrial dark by default, light theme on a toggle, persisted per browser. Respects prefers-reduced-motion.
vs. Halcon, Cognex, in-house Python, cloud vision SaaS.
NexusVisionOrchestrator is an orchestration layer, not a vision engine. It composes Halcon, Cognex, Python CV and cloud inference rather than competing with them.
| Capability | NexusVisionOrchestrator | MVTec Halcon | Cognex VisionPro | In-house Python + OpenCV | Cloud Vision SaaS |
|---|---|---|---|---|---|
| Role | Orchestration layer | Vision engine | Vision engine | Custom stack | Cloud API |
| Multi-vendor AI | ✓ native | — | — | Possible, custom | Vendor-locked |
| PLC native | OPC UA / ADS / S7 | Via add-on | Via add-on | Custom drivers | — |
| On-premise mandatory | ✓ | ✓ | ✓ | ✓ | No — cloud only |
| Visual recipe editor | Rete v2 DAG, live | HDevelop | QuickBuild | Code | Limited |
| Licensing | Per orchestrator | Per runtime | Per camera | Build cost | Per image |
| Hard real-time | Soft only | Soft | Soft | Depends | No |
Halcon, Cognex VisionPro and similar SDKs integrate as adapters — your existing investment is preserved.
Built for six teams. Pick yours.
Different stakeholders care about different things. Tabs are deep-linkable via #audience-procurement, etc.
One strategic platform
Replace many isolated point solutions with a single orchestrator. Predictable cost. One team to staff and train.
Vendor-neutral by design
Not betting the line on one AI provider. Today's engine can be tomorrow's adapter. Future engines plug in the same way.
Audit-ready out of the box
Profile versions are immutable. Every change is logged. Every inspection has a correlation ID. Compliance gets the trail it needs.
On-premise control
Your IP, your images, your servers. No per-image SaaS bill. Air-gapped networks work fine.
Vendor independence
Negotiate AI engines on merit, not on lock-in. Run several vendors side by side. Walk away when one over-prices a renewal.
Investment protection
Existing Halcon scripts, vendor SDKs and custom Python CV all integrate as adapters. Nothing has to be thrown away.
Lower integration effort
One platform replaces N×N point integrations. Add a new line by adding a profile — not a new project.
Predictable TCO
Lightweight pre-gates cut expensive per-inference license consumption. One platform team replaces many siloed integration teams.
Immutable profile versions
Sessions pin the recipe they started with. A mid-shift change never silently affects a running lot.
End-to-end correlation
One ID follows a request through REST/gRPC ingest, worker dispatch, PLC I/O and exporters. Live and historical views.
Recipes & fixtures
Save subgraph fragments + sample frames + presets as reusable blueprints. Export as CI fixtures for byte-equal regression assertions.
Ensemble & review routing
Classic CV and AI verdicts can be fused. Agreement auto-passes; disagreement is routed to human review.
Visual recipe authoring
Drag-and-drop typed DAG editor. No C#, no rebuilds. Type-safe sockets reject wrong wiring at edit time.
Live tuning panel
Drag a slider, observe the result. Pin production frames and dial parameters against real images.
Live data on every node
The editor is also the cockpit. Tag values, AI verdicts, sparklines and thumbnails — right inside each node.
Recipes that ship
Pre-AI focus gate, ROI prefetch, ensemble fusion, PLC-triggered cycle — four production patterns, all tested.
Windows Service deployment
Signed MSI installer. Config dialogs, firewall rules, service-management page. Standard IT lifecycle.
Local or LDAP auth
Three roles: Admin, Engineer, Operator. PBKDF2 local users or AD group mapping. Cookie-based session.
SQLite for state
No extra database to administer. One file to back up. Replication if you want it.
IP-based ingestion approval
New cameras queue for engineer approval before they can submit. No silent or rogue ingest.
Protocol-agnostic workers
Implement IVisionProvider. Register. Ship a model. The pool picks it up — no platform changes.
Compare engines on real traffic
Route lot A to Maddox Edge and lot B to CodingNexusVision concurrently. Or run two engines on the same image and diff verdicts.
Classic CV pre-gates
Focus score, ROI prefetch and template-match nodes upstream of inference. Cut GPU load without losing accuracy.
Live data for free
The new INodeExecutionObserver hook auto-feeds the editor's live slot. No extra wiring.
Direct answers, no marketing fog.
What is NexusVisionOrchestrator?
An on-premise AI orchestration platform for industrial image inspection. It accepts image-evaluation requests from cameras, applications or PLC triggers, routes them to the AI engine or classic CV tool best suited per request, and delivers results to databases, MES, ERP, or back to the PLC. Vendor-neutral by design. Runs as a Windows Service on .NET 8.
Which AI engines does it support?
Any engine wrappable behind a thin IVisionProvider adapter. Out of the box: Maddox Edge, Estek, the first-party CodingNexusVision engine, ONNX Runtime, custom in-house models, and more on request. Different profiles, lots, or even cameras on the same line can route to different backends simultaneously.
Which PLCs are supported?
Siemens S7-1500 and S7-1200 (via OPC UA or Sharp7), Beckhoff TwinCAT 3 (via ADS or OPC UA), and any CODESYS V3 runtime (via its built-in OPC UA server). Three drop-in flow nodes — plc:read-tag, plc:write-tag, plc:verdict — plus rising-edge trigger services with debounce.
Does it run in air-gapped factory networks?
Yes. 100% on-premise. No per-image SaaS bill. No outbound telemetry required. No cloud dependency. Deploys as a Windows Service with a signed WiX MSI installer.
How does it compare to Halcon or Cognex VisionPro?
It does not replace them. Classic CV operators run alongside AI in a single flow graph (OpenCvSharp under the hood with Halcon-parity operators). Existing scripts and vendor SDKs integrate as adapters. You can run vendor inference and classic CV on the same image in parallel and fuse the verdicts. See the comparison table.
Is hard real-time supported?
Soft real-time only. Typical jitter is ≤ 5 ms on Beckhoff ADS, 5–20 ms on OPC UA, polling-bound (default 50 ms) on Siemens Sharp7. Hard real-time and safety-PLC integration are explicit non-goals.
Who is the platform for?
Manufacturers running vision inspection across one or many lines, especially those who already use — or are evaluating — multiple AI vendors. Six stakeholder groups: management, procurement, quality, production engineers, IT, and vision/ML engineers. See Audiences.
How long does a typical pilot take?
Two to four weeks from contract to first inspection cycle running through the orchestrator in production, assuming the first AI engine is one we already adapter for. New vendor adapters typically add one to two weeks depending on SDK complexity.
How is it licensed?
Per-orchestrator on-premise license — not per image, per camera or per inference. Book a demo for current pricing.
Book a NexusVisionOrchestrator demo.
30 minutes. Live walkthrough. Engineering-led. Bring your hardest use case.
What happens after you book
- Confirmation within one business day.
- Pre-call email asking your hardest use case so the demo is targeted.
- 30 min live walkthrough. No slide deck.
Book a 30-min demo by email:
If you can see this, JavaScript is disabled. The interactive picker requires JS — the email link above works for any browser.