VisionAI
Inspection programs that return a verdict with its measurement.
What it is
A machine-vision inspection platform. An engineer builds a program, adds jobs that take an image from a camera or a file, and adds vision tasks to each job: find circle, find line, find angle, find distance, code and character reading, anomaly and classification. A run returns a pass or a fail, the per-task results, and the overlay that shows where the measurement was taken.
The backend is FastAPI with Socket.IO and SQLAlchemy over an OpenCV engine. The operator surface is a Vue 3 and Quasar application with a Konva image workspace. Updates reach every open screen over a WebSocket, so two people never see two different states of the same line.
Who it is for
Process engineers author and tune programs on the image workspace, drawing the region each task measures. Operators work from a single screen: trigger, verdict, overlay. Quality and administrators read history and manage users and settings. A single-active-operator lock means one person at a time owns the line.
How it is governed
A verdict is never a bare label. Each task reports its measurement, and the overlay shows the region it was read from, so a fail can be argued with. Programs export and import as a unit, so the thing that was running on a line is a thing that can be moved and inspected. The deep-learning tools take their models from VisionML, which pins each artifact by content hash.
What it does not do
It does not train models; that is VisionML. It does not build or sign the device image; that is VisionEdgeOS. It does not manage a fleet; that is VisionFleet. The deep-learning tools ship as deterministic stubs until the real inference backend lands. The current security posture is an MVP one, stated plainly in the repository.
Status
In development. Version 0.1.0. The roadmap records Phase 4, the controlled approval workflow, as delivered in July 2026; hardening, identity and audit-trail phases are scheduled beyond it.
Related
VisionML produces the models the deep-learning tools will run. VisionEdgeOS carries VisionAI onto an industrial device. VisionFleet reports which version of it each device is actually running.