Our platform provides a feature to define "custom models" (this is: models that may obey a particular user-defined format, provided its execution is comformant to our standard), which is documented here.

However, the platform also provides a way to deploy different versions of the model (this involves creating a new service that is meant to be accessible to the end user, which makes use of the model version to make predictions with the incoming data) in real-time and maintain related details, and a way to access that deployed API service to make predictions.

Real-time API section

In the models section, select any model and then select edit version. There you will find a tab named REAL-TIME API.


There, the user can see a list of the already defined model versions and:

To access the deployed service (for a specific version), our platforms offers a gateway which allows the user to access the corresponding resource via two different methods:

The customized path is optional and can be assigned to one version of the model (custom paths are unique among versions in the same model - when a version gains a new alias, the sibling version having it previously will automatically lose it and have no customized path until a new one is assigned to it), and later changed as will.

The custom path will only consist of alphanumeric characters, underscores and hyphens.

This feature gives the end user flexibility to deploy new versions of a model and, by using custom paths and maintaining them across versions, update the service(s) without any need to change their client scripts to hit the gateway and make predictions.