Edge AI, end-to-end
From AI Silicon to Managed Fleet — One Platform
Most edge-AI signage products are partial. Hardware vendors sell boxes; CMS vendors sell software on someone else's boxes; specialty AI players sell vertical apps. TelemetryOS ships the complete stack: validated AI silicon, the OS underneath, the runtime catalog above it, and the cloud that manages all of it as one fleet.
- On-Device LLMs
- Run open-weight mid-size language models directly on the Node Max NPU, iGPU, and CPU via llama.cpp, Ollama, and ONNX Runtime. Conversational kiosks, retrieval-augmented product discovery, on-device translation, multilingual wayfinding — without the latency, privacy exposure, or per-query cost of cloud LLMs.
- Multi-Stream Computer Vision
- Detection, classification, OCR, pose, and embedding models running on live USB or IP camera feeds for audience analytics, planogram compliance, queue measurement, occupancy, PPE, and behavior-driven content. Up to 80 TOPS combined platform AI.
- Container-Managed Workloads
- Every AI runtime ships as a Docker sidecar to the TelemetryOS application. Inference, ASR, TTS, MQTT brokers, content caches, protocol adapters — orchestrated by TelemetryOS Studio across the fleet, with signed OTA for AI container images alongside the OS and applications.
How Edge AI runs on TelemetryOS
Container-Managed AI on AMD Ryzen AI Silicon
Every AI workload ships the same way: as a Docker container sidecar to the TelemetryOS application. The Player is the user-facing screen application. The AI runtime is a sidecar exposing local OpenAI-compatible or gRPC endpoints. TelemetryOS Studio manages both as one fleet.
- AMD Ryzen AI silicon on Node Max. AMD Ryzen AI 9 HX 370 — 12-core Zen 5 CPU, Radeon 890M RDNA 3.5 GPU, and XDNA 2 NPU rated at up to 50 TOPS. Up to 80 TOPS combined platform AI. The same silicon family AMD targets for digital signage, machine vision, and HMI in their Embedded P100 series.
- Container sidecars, not bolted-on partners. BrightSign integrates AI via partner ecosystems like Quividi. TelemetryOS ships AI workloads as Docker containers managed by the same Studio that manages the screen application. One platform, one fleet, one OTA pipeline — for the runtime and the AI together.
- Validated open-source runtime catalog. llama.cpp with Vulkan acceleration, Ollama as a hosted endpoint, ONNX Runtime for vision and speech, Whisper for ASR, Piper for TTS. TelemetryOS validates the runtime configurations end-to-end so partners and customers do not build the inference stack from scratch. Specific models refresh with the ecosystem — the runtimes and the silicon are the stable interface.
- Up to 64 GB DDR5 and 8 TB NVMe. User-serviceable memory and storage that lets you load real product knowledge bases into RAG-enabled LLMs, host model weights locally on NVMe, and cache vision-pipeline state — without compromising the screen application.
Regulated and Managed
- On-edge architecture for PHI and gaming. Healthcare PHI and casino biometric data cannot transit cloud LLMs in most regulated jurisdictions. Node Max runs the LLM and vision pipeline on the device inside TelemetryOS Edge — the data never leaves the property.
- Fleet OTA for AI containers. AI container images update through the same signed OTA pipeline as firmware, OS, runtime, and applications. Staged rollouts, rollback, and remote diagnostics across the fleet.
Use cases shipping today
What Customers Actually Build with Edge AI
The use cases below are not speculative. Each is anchored in deployments that are shipping in 2026 — Posiflex's conversational ordering kiosks, Kiehl's Lift and Learn at 45+ stores, Walmart's edge shelf AI, the HIMSS 2026 healthcare front-door push, and the rapid adoption of edge AI across casino floors and sportsbooks.
- QSR Conversational Ordering
- Vision identifies items on the tray, an on-device LLM holds the conversation, the customer says 'make that a combo' and the kiosk understands. Posiflex is shipping this today on Intel Raptor Lake. Node Max runs the same workload at INT8 quality with 64 GB DDR5 — and ships the platform underneath.
- AI Lift and Learn / Product Discovery
- Customer shows the camera a product. Vision identifies the SKU. LLM with RAG over the brand catalog answers questions — 'is this good for oily skin', 'how does this compare', 'what active ingredients'. Adjacent display delivers personalized content. Kiehl's and Macy's are shipping early versions of this pattern.
- Smart Shelf + On-Shelf Signage
- Continuous monitoring of 4-8 shelf-edge cameras, SKU detection, planogram compliance scoring, and a dynamically-updating endcap display — all on one Node Max. Walmart's edge AI shelf platform documented a 47% reduction in out-of-stock incidents using equivalent architecture.
- Healthcare Digital Front Door
- Multilingual conversational LLM for wayfinding, check-in, and accessibility — on-device because PHI cannot transit cloud LLMs under HIPAA. Integrates with Epic via FHIR. Section 504 accessibility deadlines are driving healthcare orgs to refresh patient-facing technology in 2026.
- Casino Floor and Sportsbook
- Multi-display sportsbook signage with live odds, selfie age verification at entry, vision-based responsible-gambling detection, and slot personalization — with on-edge anonymization to satisfy gaming-board data residency rules. Over 80% of major operators have AI in production by 2026.
- Multi-Container Edge Compute
- Run the TelemetryOS application, an inference sidecar, an MQTT broker, a content cache, and protocol adapters concurrently. The 12-core / 24-thread Zen 5 CPU and 64 GB DDR5 ceiling carry several heavy containers without contending with display rendering.
How TelemetryOS differs
Edge AI Without the Tradeoffs
BrightSign Series 6 tops out at undisclosed-NPU computer vision around $740 USD with partner-stitched AI. NVIDIA Jetson signage boxes deliver raw TOPS but no managed platform. Intel Core Ultra edge PCs (Lenovo ThinkEdge, NEXCOM) ship hardware only. TelemetryOS + Node Max is the only product that combines all of it.
- Vertically Integrated
- Hardware (Node Max), OS (TelemetryOS Edge), runtime, SDK, CMS, fleet management, and AI runtime catalog all ship as one product. The competitive set sells boxes and asks customers to integrate the rest.
- No CUDA Lock-In
- Standard x86 / Linux / Docker / open-source ML stacks. Workloads built on Node Max run on the same code paths in cloud development and other Linux x86 environments — no proprietary inference SDK forks.
- OCuLink eGPU Expansion
- Four lanes of PCIe 4.0 to an external GPU enclosure. Scale Node Max from the integrated 50 TOPS NPU + Radeon 890M to discrete GPU class when a workload demands it. No other Node device — and no obvious competitor in the same price band — exposes external PCIe.
- Validated AI Runtime Catalog
- llama.cpp, Ollama, ONNX Runtime, Whisper, and current open-weight vision and language model families — validated end-to-end on TelemetryOS Edge with the Ryzen AI silicon. We are honest about what works in production today and what is on the roadmap.
- Regulated-Vertical Ready
- On-device inference satisfies PHI residency (healthcare), gaming-board data residency (casinos), and increasingly common edge-inference architectural requirements in finance and government. TelemetryOS Edge is the hardened OS that carries the compliance.
- Container-Native AI Delivery
- Every AI workload — inference engine, model weights, RAG index, supporting services — ships as Docker container images managed by TelemetryOS Studio. Update, roll back, monitor, and roll out AI changes across the fleet using the same pipeline as the application.
Explore Edge AI on TelemetryOS
Node Max →
The edge-AI compute appliance — 50 TOPS NPU, up to 64 GB DDR5, OCuLink eGPU expansion.
Node Pro →
Multi-screen, container-capable, with light AI workloads — $499.
IoT Integration →
MQTT, WebSerial, USB, cameras — the bidirectional I/O that AI workloads consume.
Security & Compliance →
SOC 2, hardened OS, edge-inference architecture for regulated workloads.
Bring Edge AI to Your Screens
Talk to our sales and solutions team about the Edge AI workloads, Node Max configurations, and validated runtimes for your deployment.