Skip to main content
AcuSight is a platform for running computer vision across a fleet of edge devices — and keeping it up to date. Training models is one part of that: where most edge-ML tools stop at producing a model, AcuSight also manages the devices themselves with built-in fleet and container management, it aggregates the data your fleet produces onto live dashboards, and flags when a deployed model starts drifting so you can retrain it on fresh data. One platform, from the first image on the line to a fleet in production.
New here? You don’t need to understand the whole platform to get value from it. Pick the path that matches what you’re trying to do in Choose your path below, and ignore the rest until you need it. If any term on this page is unfamiliar, the Glossary defines them all in one place.

Watch the MLOps loop in action

A short walkthrough that takes a real model from images on the line to a deployed detector — the full AcuSight MLOps loop, end to end.

The AcuSight MLOps loop

Most ML work in AcuSight follows the same five-stage loop. You can stop anywhere — collect data today, train next week — but this is the shape of an end-to-end run:
1

Collect

Provisioned edge devices stream images in, gathered into batches automatically — or upload images yourself. Each batch closes once it’s full or goes quiet.
2

Curate

Assign a batch to a project, annotate the images, split them into train/validation/test, and freeze a versioned dataset snapshot.
3

Train

Run a training job against a dataset version — pick the model architecture and size, then watch accuracy and loss metrics stream in live.
4

Deploy

Deploy the trained model to a group of devices. Each device picks up the new model on its next check-in.
5

Operate

Watch live inference on the Video page, keep an eye on container health, dashboards, and model drift — then feed what you learn back into the next round of data collection.

More than MLOps

That’s the MLOps loop — but there’s a lot more you can do in AcuSight:

Fleet-wide management

Manage every device from one place — status, settings, and the containers running on each one — individually or across a whole group.

Data aggregation & live dashboards

Data aggregation across your fleet — every device’s data rolled up into dashboards tailored to your application, like OEE across the fleet and per machine, plus MLOps views that surface model drift.

Models that adapt

When a deployed model starts drifting, that’s a signal to retrain on fresh data and promote a new champion — closing the loop instead of shipping once and hoping.

Live video & recording

Watch live inference from any camera on the Video page, and record footage for playback and export.

Integrate & automate

A REST API, live detection and event data, and the AI Co-Pilot for building on top of the platform.

Runs where you need it

Accelerated inference on Jetson, Ambarella, and Synaptics hardware — hosted, on-prem, or fully air-gapped.

Choose your path

AcuSight serves several very different audiences. Start where you fit:

Data scientist / ML practitioner

Annotate data, version datasets, train models, and deploy them.

Infrastructure & ops engineer

Deploy the platform, provision devices, and run the fleet in production.

Developer / integrator

Build on the REST API, live detection data, and the Co-Pilot over MCP.

Explorer / hobbyist

Kick the tires — get it running and see what it does.

Under the hood

AcuSight runs as a set of cooperating services — an API gateway, an MLOps data service, a time-series pipeline, a model-training service, and a fleet/container layer built on Portainer — behind a single web app. You don’t need to know any of this to use it, but if you’re curious about how the pieces fit together, see Architecture. AcuSight is hosted at acusight.io — sign in and everything above is managed for you. Self-managed and fully air-gapped on-prem installs are also available for customers who need them.

Next steps

Quickstart

Train and deploy your first model, start to finish.

Build a defect detector

Follow the full loop end to end, from raw images to a deployed model.

Core concepts

Projects, batches, datasets, models, fleets — defined in one place.

API reference

Explore the REST API, generated from the platform’s OpenAPI spec.