Deployment
Installed on your infrastructure. In a few commands.
graal Data Platform and graal Agentic Platform install with Helm on your Kubernetes or OpenShift, with no cluster administration rights and restricted pods. You choose where they run, and you stay in control of every upgrade.
Three options
Where your data already lives
The same platforms, the same charts, three ways to host them.
On-premises
Your datacenter, your Kubernetes or OpenShift, behind your firewalls. Workloads and data stay inside your network.
Private cloud
Your cloud, your account, your network rules. The platforms install on your managed cluster like any other application.
Qualified hosting provider
Deployable with a SecNumCloud-qualified or HDS-certified hosting provider. The provider holds the label for its offer; graal installs on its Kubernetes.
Architecture
What gets installed in your cluster
A control plane, an execution layer for workloads, and your existing resources. Both platforms share your company’s SSO, the same audit conventions and the same installation: the platform and your data stay with you.
Your teams · SSO
◆ Your AI agents · MCP
graal Data Platform
- Console
- REST API and MCP
- Jobs and workflows
- Notebooks
- Distributed Spark
- Model serving
graal Agentic Platform
- Console
- API and MCP door
- Agent workflows
- A sandbox per run
- Model gateway
- Tool catalog
◆ MCP → graal Data Platform
Shared foundation
- Identity · Keycloak (SSO)
- PostgreSQL
- Audit · OpenTelemetry
Your resources
- S3 storage
- LDAP / AD directory
- Image registry
- Observability
- Your AI models
Your Kubernetes cluster · on-premises, private cloud or qualified hosting
Local trial
Try the Agentic Platform on a workstation, under Docker
One command starts a demonstration project, with no cluster and no account. Without a model key, agents are simulated; with your model provider’s key, they do the work for real. The trial is for discovery: production installs with Helm.
# one workstation, Docker only$ docker compose up✓ console: http://localhost:3000 — a demonstration project is waiting for youPrerequisites
What graal needs
Standard components your platform team already knows.
- A maintained Kubernetes or OpenShift cluster
- A PostgreSQL database, yours or the one shipped with the chart
- S3-compatible object storage
- An OIDC or SAML identity provider, or an LDAP directory (Keycloak is included with the Data Platform)
- An image registry, your own for an offline installation
- A domain name and a TLS certificate
- For agents: access to the model of your choice, hosted on your side or by a provider
Operations
Operated like your other applications
High availability
Components replicated and spread across nodes: an upgrade or a lost node interrupts neither workloads nor scheduling.
Backup and restore
Metadata, configuration and encrypted secrets backed up continuously; restoration is documented and rehearsed.
Observability
Prometheus metrics and OpenTelemetry traces, sent to your own monitoring tools.
GitOps
Pinned versions, values in your repository, upgrades applied by your GitOps tool.
Offline
Installed without Internet access
The charts, the images and their signatures are mirrored into your registry. Installation and upgrades never leave your network.
# chart and images mirrored into your registry, digests and signatures included$ helm install graal oci://registry.internal/graal/graal \ --values values.yaml --set global.imageRegistry=registry.internal✓ graal installed — console: https://graal.internalYour environment
OpenShift, multiple entities, your brand
OpenShift
Images run under the arbitrary user ID that OpenShift assigns, with no security constraint to relax.
Multiple entities
Subsidiaries, departments or sites on a single installation, each with its own directory, projects and permissions.
White label
Login screen, colors and domain name in your brand, entity by entity.
Reversibility
Leaving remains possible
What you build in graal stays readable without graal.
The software is open
Open source, under the Apache 2.0 license: you can install it, use it and evolve it without us.
The code is yours
Every pipeline exports to Pandas or PySpark code that runs outside the platform.
Open formats
Apache Iceberg tables and Parquet files on your S3 storage, readable by any compatible engine.
Documented interfaces
A REST API described in OpenAPI, the MLflow API and an MCP server: your tools keep talking to your data.
See the technical specificationFAQ
Before you install
Do we need a Kubernetes cluster?
To install, yes: both platforms install on your Kubernetes or OpenShift, on-premises or with a hosting provider. If the cluster still has to be built, graal Services help you. To try it out, the Agentic Platform also runs on a workstation, under Docker.
How long does an installation take?
In the pilot, installation takes weeks 2 and 3, including the SSO connection and the first data sources.
Who applies upgrades?
You do, at the pace you choose: every release ships with its notes and is applied with Helm or your GitOps tool. With the subscription, the vendor qualifies the updates and signs the images.
What rights does graal need on the cluster?
No administration rights. The exact list of objects created and rights granted is in the technical specification.
Plan a pilot installation
Two weeks to install the platform or platforms on your infrastructure, with your platform team: the first step of the 8-week pilot.