We charge based on the number of frames in the latest state of your machine learning dataset. You can think of this as the number of "rows" in your dataset, where each row can contain (for example) an image with labels on it.

There are two types of frames we bill for:
1) The number of unique labeled frames in your latest labeled training / validation / etc. dataset stored in Aquarium.
2) The number of unique unlabeled frames that you'd like to index for embedding similarity searches / targeted data collection. This is usually data collected from your production environment that your model has run inferences on.

This means we do not charge for metadata fields attached to each example like device id, timestamp, etc. We also do not charge for model inferences on these datasets. For object detection datasets, we don't change pricing per frame. However, we reserve the right to change pricing or impose limits on how many of these you can upload if it exceeds our system's ability to scale.

Aquarium offers a suite of pretrained models that can automatically generate embeddings on your own dataset. We can also train embedding generation models on your dataset that are fine-tuned for your data domain.

Yes. Aquarium's client API allows users to upload their own model embeddings.

Yes, users can upload URLs to data hosted outside of Aquarium with a little bit of setup. All of Aquarium's functionality functions the same as long as authentication is configured correctly.

Aquarium is SOC 2 Type 2 certified. We support SSO and other enterprise-grade authentication schemes to interface with hosted data.

In addition, we have a special "Anonymous Mode" which allows users to upload metadata to Aquarium without ever exposing their raw data.