ProbeEdge profiles your AI models against real edge devices — memory fit, latency, throughput — in seconds, not hours.
Join the waitlist for early access. No spam, unsubscribe anytime.
$ probeedge profile mobilenetv2.onnx --device rpi4
Model: MobileNetV2 (ONNX)
Target: Raspberry Pi 4 · 4GB · Cortex-A72
Memory
12.4MB
of 3.7 GB free
Latency
43ms
±15%
Throughput
23inf/s
batch 1
Calibrated on native ARM hardware · confidence intervals included.
// THE PROBLEM
You won't know if your model exceeds device RAM until you've already deployed it and watched it crash.
Latency estimates from your dev machine are meaningless. ARM ≠ x86. You need real hardware numbers.
Flash an SD card. SSH in. Run inference. Fail. Tweak. Repeat. Hours lost on what should take seconds.
// HOW IT WORKS
We read only the model config — architecture, layers, and tensor shapes. Your weights never leave your machine.
Raspberry Pi 3/4/5, Jetson Nano/Orin, STM32, or custom specs.
Memory fit, latency, throughput, and OOM risk — each with a confidence interval.
Model fits// WHAT YOU GET
Run `probeedge profile` straight from your terminal. No GUI, CI/CD-ready, scriptable.
Memory fit, operator compatibility, and OOM detection from your model config alone — no device, no upload.
Pro runs your architecture on native ARM hardware. Latency and throughput measured on the metal — never extrapolated from a dev machine.
Every metric carries ±% bounds. We tell you what we're sure about — and where the numbers get fuzzy.
Profile against a growing library of edge targets — Raspberry Pi, Jetson, STM32, and more.
// PRICING
Pricing locked for early access members.
Local prices are indicative. Billing at launch is processed in USD.
We're launching soon. Early access members get Pro free for 3 months.
Your model weights never leave your machine — we only read architecture metadata.
See how we measureSupported frameworks: