Rollo

Trust & security

Your AI. Your data. Your rules.

AI only goes into production when your IT, security and legal teams can sign off. This page sets out where your AI runs, what happens to your data and how you keep control, before and after we hand over.

Request the security pack Talk to us

At a glance

Hosting
Your cloud, your data centre or Rollo-managed in the EU
Your data
Never used to train a model
Platform
Open source: kagent (CNCF), Kubernetes, MCP
Traceability
Every prompt, action and model call logged
Compliance
GDPR-compliant, ready for the EU AI Act
Exit
An exit plan in every contract

Three commitments we put in writing.

  1. 01

    Your data stays yours

    Nothing leaves without your say

    Your data is processed where you decide and never used to train a model. We work with model providers on business terms that exclude training.

  2. 02

    You choose where it runs

    Your environment, or ours in the EU

    In your own cloud tenant, in your data centre, or managed by Rollo on European infrastructure. You pick per use case.

  3. 03

    You keep the keys

    Open source, no lock-in

    The platform is open source and the agents, data and configuration are yours. Every contract includes an exit plan.

Hosting

Where your AI runs.

The same platform runs in each of these setups, so you can start in one and move to another without rebuilding.

  • Your cloud

    In your own tenant

    We deploy on Kubernetes in your Azure, AWS or Google Cloud environment, in the region you choose. Your access rules and your logs.

    Best when your cloud team already runs Kubernetes.

  • Rollo-managed

    Run by Rollo, in the EU

    We host, monitor and update the platform for you on European infrastructure, with service levels agreed in your contract.

    Best when you want results without extra IT load.

  • On-premise

    In your own data centre

    For sensitive data, the platform runs fully inside your network, with open-weight models hosted locally if needed.

    Best for regulated and confidential workloads.

Models: you choose per use case between locally hosted open-weight models, models hosted in the EU and the major providers. The AI gateway routes every request according to your policy.

Built into the platform

Security in every agent, by default.

Agents get only the access they need, and every step they take can be traced afterwards.

  • Permissions

    Each agent acts with least-privilege access to your systems, tied to your identity provider and roles.

  • AI gateway

    One policy for every model call: which data may go to which model, from which team.

  • Guardrails

    Checks on input and output, with a human approving the decisions you choose.

  • Audit trail

    Every prompt, action and model call is logged and traceable, ready for an audit.

  • Evaluations

    Agents are tested against your own cases before release and monitored after it.

  • Cost control

    Budgets and spend per agent and per team, so nothing runs up a bill unnoticed.

Compliance

Ready for your legal and risk teams.

  1. GDPR

    A data processing agreement, a list of sub-processors and model providers, and personal data minimised or pseudonymised by default.

  2. EU AI Act

    We assess each use case against the risk categories, document how it works, build in human oversight and train your people in AI literacy.

  3. Your policies

    We follow your security policies, fill in your security questionnaire and walk your team through the architecture before anything goes live.

Exit plan

Leaving should be easy. So you stay because it works.

Every contract includes an exit plan. If you stop working with us, you keep everything you need to run your AI yourself.

  • Agents, prompts and configuration in your own repositories
  • An open-source platform your IT team can run without Rollo
  • Your data, logs and evaluations exported in open formats
  • Documentation and a handover to your team
  • Your people trained to build and run AI themselves
  • No licence that stops working when the contract ends

What we share with your security team

  • Architecture walkthroughOn request
  • Data processing agreement (DPA)On request
  • Sub-processors and model providersOn request
  • Your security questionnaire, completedOn request
  • Security fact sheetIn preparation

FAQ

Questions we hear.

On model training, where data lives, model providers, access and exit.

Your question isn’t here? Send us your questions

Is our data used to train AI models?

No. Your data is never used to train a model, neither by Rollo nor by the model providers we work with on your behalf.

Where is our data stored and processed?

Where you decide: in your own environment, on-premise, or on European infrastructure managed by Rollo. The AI gateway enforces which data may reach which model.

Can we use our own model provider or contract?

Yes. The platform uses standard model interfaces, so it works with the providers and agreements you already have.

How do you handle access to our systems?

Agents get least-privilege access through your identity provider. Rollo engineers work with named accounts and the access you grant, and every action is logged.

What happens if we stop working with Rollo?

You follow the exit plan in your contract: you keep the platform, the agents, the configuration and the data, and your team can run it without us.

Bring your security team to the first call.

We walk your IT, security and legal people through the architecture and answer their questions before anything goes live.

Request the security pack Explore Rollo AI OS