Skip to content

Getting Started with Foundry🔗

Choose the customer deployment path🔗

Foundry is deployed in your organisation's environment. Foundry Hub uses Kubernetes and is packaged with Helm. Foundry Hub can be installed on an existing Kubernetes cluster, or on a single server/VM.

As an alternative to Foundry Hub, you can run standalone tasks without installing Hub, using the runtime and compute resources specified for each task.

Path What it involves
Foundry Hub on an existing Kubernetes cluster Install the Apheris Foundry Hub Helm chart and configure database, persistent storage, network access, authentication and the compute resources needed by your workflows.
Foundry Hub on a single server/VM This deployment option runs Foundry Hub on a single server.
Foundry tasks as container images Get the Foundry task container images and instructions, then run them using the container runtime and compute requirements specified for those tasks.

Agree the deployment option and infrastructure requirements with Apheris before setting up your environment.

How software reaches your deployment🔗

Foundry delivers task software as container images using the Open Container Initiative (OCI) format. Your organisation receives access to the images published to its configured repository, according to its licensed capabilities and relevant network or project access. Foundry Hub deployments and Foundry task users obtain their container images through this delivery mechanism.

A container image packages executable tasks and can also include the initial model weights they need. The checkpoints produced by a fine-tuning run stay on your side only.

Follow the instructions supplied with your Foundry version to keep your installation up to date.

Prepare infrastructure for Foundry Hub🔗

Use the installation guide supplied with your Foundry version to confirm the supported Kubernetes and Helm versions and configure the following:

  • Compute: CPU and memory for Hub and workflow tasks, plus NVIDIA GPU support for GPU workloads. Size GPUs and storage for your models, datasets and expected number of simultaneous runs. Account for these workload requirements in addition to the resources needed to run Hub.
  • Database: PostgreSQL for Hub application records. Configure the database and its permissions as described in the installation guide for your Foundry version.
  • Persistent storage: storage supporting both ReadWriteOnce and ReadWriteMany access. Shared storage holds workflow inputs, outputs and working files so the Hub, workflow launcher and task containers can access the same run data.
  • Network access: an ingress or gateway controller, a hostname and TLS configuration for Hub, and access to the required image registry and any configured services.
  • Identity and access: a configured identity provider, user access and credentials for the container images. Installation credentials and end-user sign-in are separate concerns.
  • MSA service, where needed: a configured alignment service, such as a self-hosted ColabFold service, or uploaded alignments as supported by the workflow. Select the service with your data-handling requirements in mind.

In the Kubernetes deployment, Hub starts a workflow launcher job and the workflow's tasks run as Kubernetes workloads. Persistent storage connects the stages and retains the run files. The private installation guide covers the permissions, storage configuration and operational details for this setup.

Install and verify🔗

Your platform team installs the agreed Foundry Hub version and configures its dependencies using the supplied guide. Before enabling scientific users, verify that:

  1. Users can sign in and see the expected workflow catalogue.
  2. The deployment can access its database, shared storage, licensed images and configured MSA service.
  3. A representative example workflow completes, and its status, logs and results are accessible.

For Foundry tasks as container images, use the example input and execution instructions supplied with the task to verify the container runtime, model access and compute resources.

Run your first workflow🔗

  1. Open the workflow instructions. Check the required files, model selection, parameters and expected results for your deployed version.
  2. Submit an example. Use the guided Foundry Hub screen where provided, or the Foundry Hub CLI (PyPI: apheris-foundry-cli) for command-line workflows or the Skills for agentic interaction. For OpenFold3 prediction, this can be a query manifest and its referenced assets, or supported CIF inputs.
  3. Inspect execution. Follow run status and review the logs if a step fails.
  4. Review and download results. OpenFold3 prediction returns structure files and confidence scores. Structural fine-tuning returns dataset information, metrics and checkpoints. Use the result manifest to identify the files belonging to the run.

Support and detailed guides🔗

For installation instructions, workflow examples, input and output contracts, CLI/API references or ADMET task guides, contact your Apheris representative or support@apheris.com. Include your deployment version, execution path and intended task or workflow so the team can provide the applicable instructions.