# A curated model catalog: ONNX inference on LSD
URL: https://lsd.tools/blog/onnx-model-hosting
Updated: 2026-07-17

> LSD now ships a curated catalog of hosted ONNX models you can run as a step in any pipeline. Choose a model, and its predictions flow downstream to 3D, exports, and dashboards like any other transform — each run charges a clear, fixed price to your credits.

## From chat to a curated catalog

Our first take on hosted AI was a chat assistant. It was fine — but it wasn't where the value was. What people actually wanted was to run real models against *their own* data, inside the pipeline they were already building. So we host a catalog of models you can drop straight onto the canvas — vetted, priced, and ready to run.

## Pick a model, get a step

Open the model picker in the desktop app and choose a model from the catalog. Each one is grouped into a category and shows its per-run price. Wire your features in, and the model's predictions come out as a typed table — ready to join, filter, visualize in 3D, or export.

Because it's a first-class step, everything that makes LSD pipelines reproducible applies: the output is cached, it re-runs automatically when the inputs change, and the whole path from raw data to prediction stays inspectable.

## A clear price per run

Every model in the catalog carries a fixed per-run price, shown before you run it. You pay for exactly the runs you make against your credits — no surprise metering, and nothing to pay for a model you don't run.

## Get started

Browse the [catalog](/store), drop a model into a pipeline, and run it. See the [docs](/docs) for input conventions and the models available, and [subscribe via RSS](/rss.xml) to catch what ships next.

## FAQ
### Which models can I run?

The ones in LSD's catalog. Each model is vetted and hosted by LSD, grouped into categories (image, agents, embeddings, and more), and shows its per-run price up front. You select one from the desktop app and run it as a step.

### How is inference billed?

Each run charges the model's fixed per-run price to your account credits — you pay for the runs you actually make. A disabled model can't run and never charges.

### Where does the model run?

On LSD's hosted inference service. Your pipeline calls it like any other step; the inputs and outputs are typed entities, so results are cached and re-run reactively when upstream data changes.


## References
- [Model catalog](https://lsd.tools/store)
- [LSD docs](https://lsd.tools/docs)