Skip to main content
This quickstart walks the entire AcuSight MLOps loop — from images on the line to a model running live on a device — right in your browser. There’s nothing to install.
Prefer to watch first? What is AcuSight has a short video of this exact flow. And if a term doesn’t ring a bell — batches, dataset versions, champion — the Glossary defines them all in one place.

​
Before you start

You’ll need:
  • An AcuSight account. Sign in at acusight.io. (If your organization runs AcuSight on-prem, use its URL instead.)
  • Some images to work with — either an edge device that’s streaming images, or a folder of images on your computer to upload.

​
Build your first model

1

Sign in

Go to acusight.io and sign in. You’ll land on the Home dashboard, which shows your devices and their live status.
2

Create a project and bring in images

Open Projects and create one (say, Cable glands). Then add images:
  • From a device — images your devices capture arrive automatically as batches. Open a batch and choose Move Batch to Project.
  • Upload — use Upload Data to add images straight from your computer.
3

Annotate

Open Annotate. Your batch starts under Unassigned — open it to begin labeling. Draw a box around each object and assign it a class (for example, pass or fail); you manage classes and their colors on Classes & Tags. Mark images as reviewed as you work through the batch.
4

Create a dataset version

Once enough images are labeled, add them to the dataset and pick a train / validation / test split. Then open Snapshots → Create Snapshot to freeze an immutable dataset version. You can optionally add preprocessing (such as dropping un-annotated images) and augmentation (such as horizontal flips) to the training split.
5

Train a model

Choose Train Model on your dataset version. Pick an architecture — YOLOv8 is a solid default — and a model size (Nano is fast and ideal for edge devices). Accuracy and loss metrics (mAP, precision, recall) stream in live as training runs.
6

Deploy and watch it run

Open Models, select your trained model, and choose Deploy. Pick the group to deploy to and confirm — every device in it picks up the model on its next check-in. Open Video to watch live inference — detections appear in real time, with Good / Defect events listed alongside.
That’s the full loop. Adapting a model later — as parts change or new defects appear — is the same path: collect more images, re-annotate, snapshot, retrain, redeploy.

​
Next steps

Where you go next depends on what you’re here to do:

Build models

Go deeper on annotation, dataset versioning, training, and deployment.

Run a fleet

Provision edge devices and operate them in production.

Build on the API

REST API, live detection data, and Co-Pilot/MCP.

Just explore

Follow a tutorial and see the platform end to end.
Want this same flow with a worked example and screenshots? Build a defect detector walks the whole loop as a guided tutorial.