Connectors
Databases, files, S3, SFTP, REST APIs, Hadoop and SAS, connected through a reference to an encrypted secret.
graal Data Platform
From source to model in production, on your infrastructure
From the first connected source to the model served in production, every building block of graal Data Platform shares the same identities, the same permissions, the same catalog and the same audit trail. Your teams and your AI agents work in it together. Open source, under the Apache 2.0 license.
Capabilities
Nine capabilities, one console
Each one stands on its own, and they all talk to each other: a connection feeds the pipeline, the pipeline writes the table, the table feeds the model.
Low-code pipelines
You draw, the Data Platform writes the Pandas or PySpark code, and it is yours.
Orchestration and compute
Jobs, workflows, cron and event triggers; distributed Spark and GPUs on demand.
Lakehouse and SQL
Iceberg tables on your S3, federated SQL with Trino, catalog and lineage.
Notebooks
Jupyter and VS Code in the browser, with a coding assistant wired to your own model.
Machine learning
Experiments, model registry and serving, with drift monitoring.
Generative AI
An LLM gateway and RAG over your documents, with the models you choose.
Governance
Per-project permissions, audit of humans and agents, encrypted secrets, costs and quotas.
Run by your agents
An MCP server so your agents create, run and schedule work, under your rules.
Architecture
How the building blocks fit together
A control plane, an execution layer, your own resources: everything runs in your cluster. Your teams come in through SSO, your agents through MCP, and both go through the same API.
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
One identity
Your directory and SSO for people, service accounts for agents and applications.
One permission model
Roles per project and per resource, applied the same way in the console, the API, SQL and MCP.
One catalog
The tables written by pipelines and jobs are the ones read by SQL, notebooks and models.
One audit trail
Every action, human or agent, is traced with its author, its resource and its result.
Built on open standards
- Kubernetes
- Helm
- Apache Iceberg
- Trino
- Apache Spark
- Jupyter
- MLflow
- Keycloak
- PostgreSQL
- S3
- MCP
Openness
Nothing locks you in
The code of the Data Platform is open source, under the Apache 2.0 license. What you build with it stays readable, exportable and drivable by your own tools.
Exported code
Pipelines export to Pandas or PySpark and run outside the Data Platform. Jobs and pipelines are versioned in Git.
Open formats
Parquet and Apache Iceberg on your S3 storage, standard SQL with Trino: your data can be read without the Data Platform.
Public APIs
A REST API described in OpenAPI, the MLflow API for model tracking, and MCP for agents.
Your automation tools
A Terraform provider and a command-line interface, generated from the same contract as the API.
The foundation
What holds it all together
Run by your agents
One service account per project, human approval for sensitive actions, no destructive tool enabled by default, and every call visible live.
Explore agent-driven operationsGovernance
Permissions per project and per resource, audit of humans and agents, encrypted secrets, costs and quotas.
See governanceDeployment
One Helm chart, on-premises, in a private cloud or with a qualified hosting provider, highly available and all the way to offline installation.
See deployment optionsWith the Agentic Platform
Your agents operate your data, under guarantees
graal Agentic Platform moves your processes forward with agents and people, under measured guarantees. Its agents operate the Data Platform through its MCP server, under a service account limited to one project, and a data change they propose is governed end to end.
The two platforms are independent: each can be installed, used and bought on its own.
Discover the Agentic PlatformFrequently asked questions
Do we have to adopt everything at once?
No. Many start with jobs and orchestration, or with the lakehouse, then expand. Each capability works on its own and joins the others without migration.
Where does processing run?
In your Kubernetes or OpenShift cluster, on-premises, in a private cloud or with the hosting provider of your choice. Data stays in your storage.
Can we keep our current tools?
Yes. Your Python and Spark scripts, dbt projects and MLflow clients work as they are, and your Iceberg tables stay readable by any compatible engine.
Which model drives the agents and generative AI?
The one you choose: a model hosted on your premises and served by the Data Platform’s LLM gateway, or the provider of your choice.
Do we need the Agentic Platform too?
No. graal Data Platform and graal Agentic Platform are independent: each can be installed, used and bought on its own. The Agentic Platform’s agents can operate the Data Platform through its MCP server.
Let's see the Data Platform on your data
An end-to-end demonstration, from source to served model, driven by your teams and by an agent.