Modes of Deployment

Truefoundry provides a split plane architecture which comprises of a control plane and a compute plane. The control-plane orchestrates the deployments across multiple compute plane Kubernetes clusters connected to the control plane. You can read more about the Truefoundry architecture here: https://blog.truefoundry.com/truefoundry-ml-platform-on-kubernetes/.

Because of this, Truefoundry can be deployed in two ways on your own cloud:

Compute-plane in your own cloud environment and use hosted Truefoundry control plane

Truefoundry hosts a multi-tenant control plane which can orchestrate deployments across multiple compute-planes. In this case, you create a Kubernetes cluster in your own cloud account and connect it to the hosted control plane. The connection is done via installing the tfy-agent in your own kubernetes cluster which helps it connect to the hosted control plane. In this case, all compute and data stays in your own cloud environment - however the control plane has access to fetch the artifacts and logs of your applications.

This is a more managed setup wherein the upgrade of Truefoundry is automatically handled by the Truefoundry team.

To understand all the steps involved in connecting cluster to control plane, you can read Deploy Compute Plane. To get started with deploying, follow the guide below based on your cloud provider.

Both compute and control plane in your own cloud environments (Enterprise Plan)

In this case, both the control plane and the compute plane are hosted on your own cloud environment. The control plane comprises of the following components:

Truefoundry helm chart (This can be installed on any Kubernetes cluster - including one of the compute plane clusters itself.)
Postgres Database
One Blob Storage (S3 bucket / Azure Container / Google Storage Bucket / Minio)

To get started with installing the control plane and compute plane, please follow the guide: Deploy Control Plane and Compute Plane