# ONNX Runner
URL: https://lsd.tools/store/onnx-runner
Updated: 2026-07-17

> Drop an ONNX model into your pipeline as a step. Wire its inputs from upstream entities, run inference with batching, and feed predictions downstream — every result cached and inspectable.

## What it does

ONNX Runner makes a trained model just another node on the canvas:

- **Wire inputs** from any upstream entity (a feature table, an image set).
- **Batch or single** inference with cached outputs.
- **Overlay predictions** back onto the source for inspection.

## Typical pipeline

Shape your features upstream, connect them to the ONNX Runner node, choose the model file, and route predictions to a delivery step or a 3D overlay.

## Install

Click **Open in LSD** to install. The ONNX Runner step appears in the palette under Machine Learning.

## FAQ
### What model formats are supported?

Any standard ONNX model. Inputs are mapped from upstream entities and outputs become entities you can inspect or feed onward.

### Does inference run locally or hosted?

ONNX Runner executes the model as a pipeline step. For hosted, billed inference of larger models, see the Models page.


## References
- [Machine Learning use case](https://lsd.tools/use-cases/machine-learning)