Quickstart

Install the agent, connect a camera, label 50 defect images, train overnight, and activate a model that sends a pass/fail reject signal to your PLC. This guide covers the full path. Expected time to first live inspection: under 4 hours of active work, plus overnight training.

Prerequisites

Before you begin, ensure you have:

• Industrial PC: Intel Core i7 10th gen or above, 16 GB RAM, Ubuntu 22.04 or Windows Server 2019+
• Line camera connected via GigE Vision, USB3 Vision, or RTSP
• Network access to camera from IPC (same subnet or routed)
• Your Procunit license key (emailed after pilot approval)
• 50+ labeled defect images, or access to your line to capture them

Install the Procunit Agent

Run the installer script on your IPC. The agent pulls the ONNX runtime and Labeler UI.

# Ubuntu 22.04
curl -fsSL https://install.procunit.com/agent | bash
# Windows Server (PowerShell, run as Administrator)
iwr https://install.procunit.com/agent.ps1 | iex

After install, activate with your license key:

procunit activate --key YOUR_LICENSE_KEY

Connect a Camera

Edit the agent config at /etc/procunit/agent.yaml:

cameras:
  - id: line-1-cam
    protocol: gige
    address: "192.168.1.101"
    resolution: "1280x1024"
    fps: 30
    trigger: hardware

plc:
  interface: ethernet_ip
  address: "192.168.1.50"
  reject_coil: "O:0/1"

Verify the camera connection:

procunit camera test --id line-1-cam
# Output: camera=line-1-cam status=OK fps=30 resolution=1280x1024

Label Defects

Open Procunit Labeler in your browser at http://localhost:8090. Capture or upload 50+ images containing defects. Draw bounding boxes around each defect region.

Labeler keyboard shortcuts: B draw box — Del delete selection — Enter confirm + next image.

Trigger Training

Once labeling is complete, start a training run. Training takes 6-12 hours depending on dataset size. Run it overnight.

procunit train --dataset line-1-defects-v1 --model-name door-latch-v1
# Training starts in background. Monitor with:
procunit train status

When training completes, activate the model on your line:

procunit model activate --name door-latch-v1 --camera line-1-cam
# Output: model=door-latch-v1 status=live latency_p95=7.3ms

Configuration Reference

Full agent.yaml schema and all config options are documented in the API Reference. Key options: