Skip to main content
AcuSight runs on the edge as a lightweight containerized agent, so the hardware bar is low: if a device can run Docker and reach the internet outbound, it can join a fleet. Beyond that baseline, AcuSight is validated on the accelerated platforms below for on-device inference.

​
Requirements

Any edge device that meets these can run the agent:
  • 64-bit Linux — arm64, amd64, or arm.
  • Docker installed, with the agent able to reach the Docker daemon (use sudo if the device user can’t).
  • Outbound HTTPS to AcuSight. No public address, open ports, or inbound firewall rules — devices connect outbound and work behind NAT.
  • A camera or image source for anything that captures data.
That’s enough to bring a device online, point it at a project, and collect data. Running a model on-device benefits from hardware acceleration — covered next.

​
Validated platforms

These platforms are validated for on-device inference with the AcuSight edge agent:

Board bring-up

Some accelerated boards need their vendor’s base image in place before the agent can be installed — prerequisites, including the Synaptics Astra setup guides.

Model export handled for you

Accelerated inference uses each platform’s runtime — a Jetson GPU, a Synaptics NPU. AcuSight exports your trained model to the right format when you deploy.
Don’t see your board? A generic amd64 or arm64 host that meets the requirements still runs the agent and inference on CPU — reach out if you need a specific accelerator validated.

​
Next steps

Edge agent

What the agent is and how it runs on a device.

Air-gapped deployments

Run AcuSight fully disconnected, on your own infrastructure.

Board bring-up

Get an accelerated board flashed and online first.

Connect a device

Bring one of these devices online.