Sovereign data & AI · open source
graal publishes two open source platforms, independent and complementary: the Data Platform, from source to model in production, and the Agentic Platform, where your AI agents move your processes forward under measured guarantees. Both install on your infrastructure.
Demo
A data change, from request to production: the agent works in its sandbox, the guarantees check, a person approves, the Data Platform applies — and everything is evidenced.
DATA-142 · add the feels-like temperature to the forecast
graal Agentic Platform· “data change” workflow
graal Data Platform· what happens there
Evidence
Animated illustration of a demo scenario, not a screenshot: ticket, people, costs and identifiers are fictitious.
The jobs of a project in the Data Platform, their status and their last run. Rows marked “Agent” were created by an AI agent through MCP.
Real graal console interface, not retouched; fictional demonstration data (tenant energie-demo — people, projects and tokens are invented).
Service account agent-previsions · project previsions
Built on open standards
Two products
Take one, the other or both: each installs and runs on its own, under the Apache 2.0 license.
graal Data Platform
graal Agentic Platform
Together
The Agentic Platform’s agents operate the Data Platform through its MCP server, under a service account scoped to one project; every call is traced on both sides.
Branch, guarantees (tests run, data quality on a sample, measured facts), human approval, application through the Data Platform API, evidence attached to the ticket.
Your company SSO (OIDC, SAML, LDAP), the same audit conventions, the same installation: Helm, no cluster-admin rights, restricted pods, on your infrastructure.
graal Data Platform
From the first connected source to a model served in production, without assembling ten tools.
Your databases, files, APIs and Hadoop estate, connected in minutes.
Draw a pipeline: the Data Platform writes the Pandas or PySpark code, and it belongs to you.
Jobs, workflows, cron and event triggers; distributed Spark and GPUs on demand.
Iceberg tables on your S3 storage, federated SQL with Trino, catalog and lineage.
Jupyter, VS Code and RStudio in the browser, with a coding assistant wired to your own model.
Experiments, model registry and serving; an LLM gateway and RAG over your documents.
An MCP server so your agents create, run and schedule work, under your rules.
Per-project permissions, an audit trail of human and agent actions, encrypted secrets, costs and quotas.
Open source, exportable code, open formats, REST API, MLflow and MCP: nothing locks you in.
How it works
One source, one secret, and it's wired.
You draw, the Data Platform writes the code.
The low-code editor canvas: every block carries its parameters, and the graph becomes code.
Real graal console interface, not retouched; fictional demonstration data (tenant energie-demo — people, projects and tokens are invented).
Schedule without writing a line of YAML.
A scheduled workflow: its steps, their dependencies and their retries.
Real graal console interface, not retouched; fictional demonstration data (tenant energie-demo — people, projects and tokens are invented).
Notebooks and SQL on the same tables.
Every training run is tracked.
From model to endpoint, in one click.
Who did what, and what it costs.
Roles: what everyone can do, and on what.
Real graal console interface, not retouched; fictional demonstration data (tenant energie-demo — people, projects and tokens are invented).
graal Agentic Platform
Software delivery, operations, joiners and leavers, data changes: describe the process, hand steps to agents from any vendor, and keep the decision. by default, nothing is written without a guarantee and a human approval.
States, transitions and actors: agents, people with a deadline and an escalation, the platform. An inconsistent workflow is refused as soon as it is written.
Claude Code, Codex, Gemini CLI or your own, behind a single protocol; each one registered, versioned, budgeted and revocable.
Tests actually run, scope respected, no secret, signed provenance: a guarantee that cannot check refuses.
By default, every write to a third-party system waits for another person’s approval, re-authenticated. No decision goes through MCP.
The model of your choice, a capped key per run, every model or tool call counted; the agent never holds a tool credential.
What the agent did is measured by the platform, logged and attached to the ticket: tests, facts, cost, approvals.
You choose the model: hosted on your premises, or with the provider of your choice. Every run works in a disposable sandbox, with the network closed by default.
Discover the Agentic PlatformThe Data Platform, run by your agents
The Data Platform exposes an MCP server. The Agentic Platform’s agents, Claude or any compatible client create, run and schedule jobs through a service account scoped to one project, and every action shows up live in the console.
You choose the model: hosted on your premises, or with the provider of your choice.
Service account agent-previsions · project previsions
The Data Platform “Agents” panel: who is connected, with which permissions, and every tool call live.
Real graal console interface, not retouched; fictional demonstration data (tenant energie-demo — people, projects and tokens are invented).
Deployment
One Helm chart per platform, no cluster-admin rights and no custom resources: graal runs where your data lives.
Your teams · SSO
◆ Your AI agents · MCP
graal Data Platform
graal Agentic Platform
◆ MCP → graal Data Platform
Shared foundation
Your resources
Your Kubernetes cluster · on-premises, private cloud or qualified hosting
Your datacenter, your Kubernetes or OpenShift.
Your cloud, your account, your network rules.
Deployable with a SecNumCloud-qualified or HDS-certified hosting provider.
Open source
Both platforms are free software: you can install, study, modify and redistribute them. The vendor subscription adds what production requires.
The Apache 2.0 license allows any use, commercial included, royalty-free.
Open formats (Parquet, Iceberg), standards (SQL, MCP, OpenTelemetry), exportable pipeline code: you leave with your work.
The vendor subscription: support, security fixes, qualified updates and signed images, per vCPU, unlimited users.
Use cases
Energy
Temperatures and regional consumption in a thermosensitivity model, refreshed every morning.
Industry
Anticipate failures from sensor data, and alert before downtime.
Public sector
Query your internal documents with a model hosted on your premises.
Insurance
Flag unusual cases and explain every score.
Research
Turn a notebook into a scheduled, governed and monitored job.
Cross-industry
Agents that watch runs, diagnose and propose a fix, approved before it is applied.
Cross-industry
Move from Saagie, Hadoop or a SaaS by bringing your Python and Spark jobs along.
Who it's for
Two open source platforms instead of ten building blocks, on your infrastructure, at a predictable cost.
See pricingYour data scientists explore, your data engineers industrialize, your agents take their share of the work.
See the Data PlatformNo rights on the cluster, restricted pods, per-project permissions — and agents that hold no tool credential.
See securityData and processing stay within your infrastructure.
See sovereigntyOne Helm chart per platform, PostgreSQL, Keycloak and S3. Nothing else.
See the architectureTrust
Sovereign, strictly defined
Get started
One use case end to end, on your data, in your infrastructure, with the Data Platform, the Agentic Platform or both. Then you decide, on evidence.
Week 1
Scoping: one use case, clear success criteria
Weeks 2–3
Installation on your infrastructure
Weeks 4–7
The use case end to end, on your data
Week 8
Review and decision
No. Each installs and runs on its own. Together, the Agentic Platform’s agents operate the Data Platform through MCP, under your rules.
Yes: both platforms are open source, under the Apache 2.0 license. The vendor subscription, per vCPU with no usage-based billing, brings support, security fixes, qualified updates and signed images.
For production, yes: both platforms install on your Kubernetes or OpenShift, on-premises or with a hosting provider. The Agentic Platform can also be tried on a workstation, with Docker. If the cluster still has to be built, we help you.
Claude Code, Codex, Gemini CLI or your own agents; and the model you choose: hosted on your premises, served by graal's gateway, or with the provider of your choice. graal speaks MCP, an open protocol.
Those labels qualify hosting offers, not software. graal deploys with a SecNumCloud-qualified or HDS-certified hosting provider, just as it does in your own datacenter.
No. The platforms and the data stay with you. If you connect an external model, it only receives what the tools it calls return.
Yes. Both platforms install from your own image registry; with a model hosted on your premises, your agents work without any Internet access.
The vendor, as part of the subscription. graal Services cover integration, migration, training and operations.
The code is open, pipeline code is exportable and formats are open: you leave with your work.
Data teams
Watch the demo (14 minutes)CIOs and CISOs
Get the architecture packPublic buyers
How to buy graal