A flow (or workflow) describes how your data flows through our system. All you need to do is tell the platform where the data comes from (e.g. an image uploaded to a Dropbox folder) and what to do with a prediction on this data (e.g. make a new entry in an Airtable base).

A model is what makes your system intelligent, your "thinking" block if you will. It is trained using the data you are using as part of the workflow.

Whenever a model makes a prediction on your data, that counts as a run. Some examples:

• Flow execution with one model = 1 run
• Flow execution with four models = 4 runs
• Calling a model through our API directly = 1 run

Data requirements depend on the task you intend to perform and the data you need to process. We tune our systems in such a way that you can get meaningful results with the least possible amount of data – usually less than you might think.

You only pay when you get value from the system. That means that for every prediction you need to manually correct, you get one bonus run in credits. Over time, models will improve based on your corrections.

Sufficiently often. We want you to be able to do what is needed in order to get reasonable results which is why we don't set a hard quota on training runs. Is this always going to be this way? Of course not. But we know that you have better things to do than training a deep learning model every day so we trust in you.

No. Our servers run exclusively in the cloud, ensuring best-in-class performance and data security.