Not a Better Camera.
A Different Architecture.

Frames Are the Wrong Primitive

Almost every machine that sees uses a frame camera: every pixel, every frame, on a fixed clock, whether or not anything changed.

A single frame captures a flat, frozen instant. Depth takes more frames: a second view to triangulate, or a sequence of projected patterns to resolve. Change takes more frames too, since motion only appears by comparing one snapshot to the next. Everything a frame system knows, it infers by stacking snapshots, and every snapshot costs time. Fast, or precise. Never both.

Worse, between snapshots it is blind. The moments that matter, contact, a slip, a sudden stop, can happen in the gaps. And in this paradigm, precision is something you buy in hardware: finer lenses, denser sensors, more projected patterns. Spend enough and you get high precision and high data density, of a frozen scene. Faster chips and better glass don't change the rule. Frames are the wrong primitive for a world that never stops moving.

How Kortx Works

Stereo event sensors sense change, continuously. Each pixel fires the instant its patch of the scene changes, within millionths of a second, and stays quiet otherwise. No shutter, no clock, no snapshots to stack. The same principle biological vision runs on: the retina fires on change, not on a schedule. Two sensors in stereo, so every change lands in 3D.

A scanning laser creates the signal. A sensor that fires only on change would go blind to a part sitting still. So Kortx does not wait for the scene to change. It changes the scene: the laser writes points across it, and the sensors catch each one the instant it lands. Moving part or still part, Kortx measures precise 3D: a real marked point in space, not depth inferred from a pattern. Other technologies project a coarse pattern of dots only to give their algorithm something to match, then infer depth, and only where a part's shape gives it away. Kortx measures distance precisely.

Software makes the precision. Precision lives in the software, not in a precisely built sensor. Kortx does not need expensive lenses or a perfect optical build. The software provides the precision, so the hardware can focus on robustness: surviving the real world.

One Stream, Two Reads

Because the data is a continuous stream, not a fixed frame, Kortx reads it two ways at once. Reading a few milliseconds of the stream tracks a moving part and guides a robot as it moves. Reading about a second of the same stream builds a dense, precise 3D point cloud, detailed enough to inspect that same part for defects and measurement.

The mechanism is simply how much of the stream you read. A few tens of milliseconds is enough to know where the part is right now: sparse points, fresh enough to steer a robot through motion. About a second of the same stream accumulates into a dense model of the part, complete enough to measure and inspect. Same data, two windows.

Tracking is what keeps the accumulation sharp. Because Kortx knows the part's position at every instant, each new measurement lands on the part exactly where it belongs, even while the part moves. The model builds in the part's own frame of reference, not the world's, so motion doesn't blur it.

The old choice between fast and precise disappears: one system does both, in a single pass.

The Physics Layer for Physical AI

Today's models learn the physical world from video: frames again, flattened further. Video shows what things do, but flat and frame by frame. The contact, the give of a surface, the speed of a fall: the physics is missing, because it lives in the gaps frames throw away.

Kortx captures exactly that layer: what moved, when, in precise 3D. The same stream that lets a robot act in the moment teaches a model what physics looks like.

The Boundary of Automation Is Moving.

18 patents, 14 granted, covering laser-event 3D sensing, the streaming data format, and the hardware-software architecture behind it.

The technology makes more sense in motion than on a page. Talk to us and we'll show you.