Your data. Your models. Your future — inside your own cloud boundary.

Poolside is a secure, enterprise-grade AI platform for software engineering, designed to run entirely within your infrastructure. This repository publishes opinionated reference infrastructure-as-code for deploying Poolside in your own cloud, a transparent, production-grade happy path for deploying the Poolside platform and models into your cloud account.

• Learn about the platform: poolside.ai
• Product documentation: docs.poolside.ai

These reference architectures include the public infrastructure and deployment patterns for running Poolside in your environment.

To deploy it, you will also need the Poolside deployment bundle, which includes:

• private Helm charts
• OCI images
• versioned release artifacts
• deployment guidance and support materials

Request access to the Poolside deployment bundle: Contact Sales

This public repository is intended to show the reference architectures, infrastructure patterns, and deployment workflow. The Poolside software artifacts themselves are distributed separately to approved customers and evaluators.

A complete deployment of Poolside inference on Amazon EKS, provisioned from a single Terraform root:

• Composable Terraform modules for the VPC, EKS cluster, CPU and GPU node groups, IAM/IRSA roles, KMS keys, the ALB ingress, the NVIDIA GPU Operator, ECR image push, the model-checkpoint bucket and uploader, and API-key secret sync.
• One runnable example root, inference: the first apply builds the infrastructure, pushes the bundle images to ECR, and uploads model checkpoints to S3; enabling install_inference_stack and re-applying installs the inference Helm release, which serves models such as Laguna on atlas.
• Quickstarts and an architecture walkthrough under aws/docs/.

Start at aws/README.md.

• The Poolside deployment bundle — contact Poolside sales to obtain access.
• An AWS account you control, with admin-equivalent IAM permissions. The deployment creates IAM roles, an OIDC provider, and KMS keys, so a read-only or restricted profile fails partway through.
• terraform (>= 1.5.7), skopeo, python3 with boto3 (for the default managed model upload), the aws CLI, and kubectl installed locally. Helm is not required; the inference chart is applied through the Terraform Helm provider. See aws/docs/prerequisites.md for versions and the full checklist.

reference_architectures/
└── aws/             # AWS reference architecture
    ├── docs/        # AWS-specific guides and quickstarts
    ├── examples/    # end-to-end Terraform roots
    └── modules/     # composable Terraform modules

• Deployment questions on this repo: open a GitHub issue.
• Product, licensing, and access to Poolside artifacts: contact Poolside sales.