Use case · Cross-industry
The operations of your workloads, handed to governed agents
graal Agentic Platform puts agents to work on graal Data Platform: they read runs, logs and metrics, and propose the useful actions. They operate through the Data Platform’s MCP server, under a service account scoped to one project; whatever changes production waits for a person’s approval; every call is traced on both sides.
The Agents panel: every connected agent, its permissions, and each of its actions, live.
Real graal console interface, not retouched; fictional demonstration data (tenant energie-demo — people, projects and tokens are invented).
The challenge
Busy on-call rotations, agents to keep in check
- Dozens of runs every night, and anomalies that wait until morning.
- Repetitive diagnoses: a late file, a column that changes type, a quota reached.
- Promising AI agents that cannot be left to act with administrator rights.
- The need to know, afterwards, who did what, on both sides.
How graal handles it
An operations agent, under guarantees
Step 01
Define the work
An operations workflow lays out the steps — watch, diagnose, propose, get approval, apply — and names the people who approve.
Step 02
Set the rules
The Agentic Platform runs the agent — Claude Code, Codex, Gemini CLI or another — in a disposable sandbox, with no tool credential and a capped model key. A policy sets the allowed operations, projects and time windows.
Step 03
Diagnose
Through the Data Platform’s MCP server, the agent reads the logs, events and metrics of a run, and writes a diagnosis the team can read. Its service account is scoped to one project.
Step 04
Propose and get approval
Routine actions — restart a job, fix a parameter, reschedule — run according to your policy; the others become proposals that someone else, re-authenticated, approves. No delete tool is enabled by default.
Step 05
Trace and revoke
Every call shows up live in the Data Platform console and stays in the audit trail, on both sides. Revoking the service account cuts the agent off.
The typical scenario: a data team of a few people, a fleet of nightly jobs, and an on-call rotation that should only be woken up when it matters. You start read-only — the agent diagnoses and suggests — then open up, one action at a time, what it may do on its own.
The two platforms stay independent: the Data Platform works with any MCP client, and the Agentic Platform also carries processes that involve no data, such as joiners and leavers or software delivery.
On 09/12/2025, Anthropic handed the MCP protocol over to the Agentic AI Foundation, a directed fund under the Linux Foundation.
What you get
- Anomalies diagnosed before the team arrives
- Routine actions done without intervention, the others submitted for approval
- A log of every agent action, readable and exportable, on both sides
- Agent permissions you widen at your own pace
What stays with you
- The Data Platform’s MCP server and the service accounts
- The workflows, policies, approvals and audit trail
- The choice of the model that drives the agent: hosted on your premises or with the provider of your choice
Capabilities involved
graal Data Platform
From source to model in production
- Run by your agents
MCP server, per-project service accounts, policies and approval.
- Orchestration and compute
Runs, live logs and metrics.
- Governance
Audit of humans and agents, revocation.
- Generative AI
A model served on your premises to drive the agent.
graal Agentic Platform
Your AI agents move your processes forward
- Agent workflows
Operations described as a process, with the people who approve.
- Guarantees and approvals
Disposable sandbox, closed network, re-authenticated approval before any write.
- Models, tools and budgets
A capped model key per run, every call counted.
- Evidence and audit
What the agent did, measured and traced.
Frequently asked questions
Can the agent delete data?
Not by default: no delete tool is enabled. You decide whether to open one, project by project, and you can require human approval for it.
Which model drives the agent?
The one you choose, project by project: a model served on your premises by the gateway, or an external provider. Every run gets a capped key. An external model receives what the agent reads during the run; the Data Platform’s tools only return what the service account is allowed to read.
Can we start read-only?
Yes, and it is the recommended starting point: the agent diagnoses and suggests, your teams act.
How do we stop an agent?
By revoking its service account on the Data Platform: its next calls are refused. The run that drives it stops from the Agentic Platform console: the ticket is stopped, its run cancelled and its model key revoked.
Do we need both platforms for an agent to operate the Data Platform?
No. Each one works on its own: the Data Platform can be driven by any compatible MCP client, under a service account scoped to one project. The Agentic Platform adds work organized as a workflow, the disposable sandbox, per-run budgets, approvals and evidence.
Hand a first on-call duty to an agent
In 8 weeks, a read-only agent, then the actions you choose to open to it.