Why we built LSD
Data work today is scattered across notebooks, throwaway scripts, and disconnected tools. Six months later, nobody — including the author — can say which cell produced which figure. We built LSD to make the whole pipeline visible: one canvas where data flows from Load to Shape to Deliver, and every transform stays inspectable.
Load · Shape · Deliver
- Load brings data in as typed entities — tables, images, point clouds, meshes.
- Shape transforms it with steps that cache their output and re-run automatically when inputs change.
- Deliver turns the result into a 3D scene, an exported file, or a published dashboard.
Because the graph is reactive, editing any upstream step updates everything downstream — no manual re-runs, no drift.
Built to be extended
LSD is a platform, not just an app. With the lsdtools Python SDK you can write custom steps, UI panels, and whole packages, then publish them to the Tool Store for others to install. Models run as steps too — wire an ONNX model into the graph and feed its predictions downstream.
Get started
Download LSD, open the docs, and build your first pipeline. We'll be posting deep-dives, tutorials, and model spotlights here — subscribe via RSS to follow along.