Introduction

TrueFoundry is a platform on Kubernetes that makes it really easy to build, track and deploy models without having a detailed understanding of Kubernetes. Truefoundry is deployed on a cluster on your own cloud so that the data never leaves your environment and you don't incur any data egress costs.

They key design principles behind Truefoundry are:

1. Cloud Native: Truefoundry runs on Kubernetes - so can work on almost all cloud providers or even on-prem.
2. ML Inherits the same SRE principles as the rest of the infrastructure within your org: Truefoundry seamlessly integrates with the rest of your software stack to provide the same SRE, security and cost optimization features to the ML Team,.
3. No Vendor Locking: We are developers and we understand that nobody wants to be vendor-locked in to a specific provider. We have put in a lot of effort to make sure that its really easy to migrate off of Truefoundry in case you need to do so. The Truefoundry APIs do not interfere your main code and we also expose to you all the kubernetes manifests we generate - so if you ever want to migrate, you can take those manifests and apply it yourself.

Key features

1. Jupyter Notebooks: Start Notebooks or VSCode Server on Cloud with auto shutdown.
2. Deploy a training or batch inference job: Write your python script, log your metrics, models and artifacts and trigger jobs either manually or on a schedule.
3. Deploy your models as APIs: Deploy the model artifact directly to get APIs or wrap it in FastAPI, Flask to host the APIs. The deployments support autoscaling, canary deployments out of the box.
4. Easy Debugging: View logs, metrics and cost optimization insights for the all services.
5. Model Registry: Track all the models and their versions in your organization along with their current deployment status, and metadata.
6. Deploy and Finetune LLMs: Deploy the open source LLMs in one click and finetune them on your own data.
7. Deploy common ML software Deploy the most commonly used ML software like LabelStudio, Helm Charts etc.
8. Manage multiple environments and promotion: Manage multiple kubernetes clusters from different environments and move workloads across them in a single click.

Truefoundry integrates with the rest of your infrastructure in the following way:

Truefoundry Architecture

Truefoundry follows a split-plane architecture that enables on-prem deployment as well as makes sure that the reliability of your services do you depend on Truefoundry.

To read more on the architecture, you can refer to this blog: https://blog.truefoundry.com/truefoundry-ml-platform-on-kubernetes/

You can connect your own cluster to the Truefoundry hosted Control plane. Our enterprise plan also allows you to host the control plane in your own account.