Plug-in on-device USB-type AI accelerator
MLD-R1 AI USB
Turn your device into an AI inference machine
One stick delivers 10 TOPS of on-device AI inference over a USB port. No chassis swap required.
.jpg)
AI Frameworks (via SDK qb)
-
Keras
-
TensorFlow / TensorFlowLite
-
PyTorch
-
ONNX
Operating Systems*
-
Windows 11
-
Linux 22.04.5
*To be updated
Host Architectures
-
x86
-
ARM
Keep your existing pipeline
Whether you're deploying a PyTorch model on a Windows laptop or running TensorFlow inference on an ARM-based industrial PC, MLD-R1 connects and runs.
Powered by the ultra-low-power REGULUS AI SoC
3 W power draw, engineered for 24/7 operation
Accelerate vision to generative AI workload at the edge
Mobilint’s NPU architecture has been validated with over 490 open- and closed-sourced models.
Product specs
NPU
REGULUS AI SoC
Performance
10 TOPS (INT8)
CPU
Quad-core ARM Cortex-A53, up to 1.2 GHz
TDP
3 W
Memory
4 GB / 8 GB LPDDR4X at 4,266 MT/s
Supported OS
Windows, Linux
Supported Frameworks
Keras, TensorFlow, TensorFlow Lite, PyTorch, ONNX
LED Indicators
1 (Power LED)
Host Architectures
x86, ARM
Ingress Protection
IP65 (TBD)
Data Interface
USB 3.1 Gen1
Dimensions (mm)
90 ✕ 40 ✕ 16
Weight (g)
88
Secure On-Prem Processing
ID verification, audit analysis, document summarization for data that cannot leave the device.
Offline Field Operations
Inspect remote infrastructure, generate reports and analyze data on a laptop. No connectivity needed.
Classified Networks
AI inference on closed defense, law enforcement and government networks for classified on-premises workload.
Factory Floor Retrofit
Defect detection, quality analysis and line monitoring on legacy terminals without hardware replacement.
Warehouse Automation
Inventory recognition, safety monitoring and logistics optimization on existing systems, offline.
Kiosks and Retail
Customer interaction, product recognition and local language processing at the point of service.
Use Cases
Speed up model validation and edge prototyping on your existing machine, wherever you are
Run models on portable diagnostic devices without network connectivity.