Security & sovereignty

Know where every agent and model call runs.

Start your security review with the data flow. It depends on where the agent runs, what it can access and which model it calls.

Forward this page to the person who needs to sign off. If it does not cover your situation, ask us directly. We would rather answer before procurement starts.

The short version

You choose the hosting boundary

Run in your cloud, your data centre or a Koru-managed environment. The location of runtime records follows that choice.

You choose the model

Run open-weight models on your hardware or use a hosted provider. Provider retention and training terms are reviewed for your selected endpoint.

You own the built agents

The agents Koru builds for you are yours. Koru Runtime and ongoing operating terms are defined separately.

In detail

Controls you can review

Where your data is processed

For customer-hosted deployments, agents run in your cloud account or data centre. For Koru-managed deployments, they run in the environment we operate for you. We document which data is processed in each location.

Customer-hosted agents use scoped credentials to reach your systems directly
Koru-managed agents process the agreed data in the managed environment
In your cloud, the runtime, traces and indexes stay in your VPC
Air-gapped deployments run without outbound connectivity

What happens to your data at the model

The selected model determines whether a request crosses the deployment boundary. We review provider terms and the data sent in each model call with you.

Run an open-weight model inside your network when model calls cannot leave it
Hosted providers receive the data needed for the model call under their agreed terms
Provider retention and training commitments depend on the selected endpoint and agreement
Credential custody is defined for the selected deployment

Who can do what

Set access rules before connecting live systems. The runtime uses role-based access and single sign-on from the first deployment.

RBAC over agents, tools, knowledge collections and configuration
SSO and SAML against your existing identity provider
Agents inherit the permissions of the identity they act as. They cannot read past them
Per-tool human approval gates, with unattended runs failing closed rather than waiting

What gets recorded

Every run leaves a durable trace. The trace location and retention settings follow the deployment and your agreed operating policy.

Append-only audit log of every run, tool call, approval, denial and config change
Step-by-step execution trace with inputs, outputs and the model used
Exportable for your own SIEM and retention tooling
Retention periods and log access are agreed for the deployment

Encryption and key management

The model never receives your credentials.

TLS for data in transit
Encryption at rest, using your own KMS keys where you want them
Secrets stay encrypted and are added when tools run. The model never receives them
No shared credentials across customers in managed deployments

How we work during the build

We agree on access boundaries for discovery, build and production before connecting systems.

Customer-hosted work uses credentials your team issues and can revoke
Production access is scoped to the support model you select
The agent and its integrations are documented for your team
Access changes are part of the deployment and support plan

Compliance

We work inside your compliance programme.

In your own cloud or data centre, the agent uses your network and access controls. Your team still reviews the new workload and any external model calls. For a deployment we manage, we review the controls and data flow with you.

We provide an architecture diagram and samples of access records and audit logs for your review. We work through residency and policy requirements before you choose a deployment.

If you have a specific framework, questionnaire or residency mandate to satisfy, send it over early. Tell us what you need to clear →

Where it runs Your cloud, your data centre including air-gapped, or managed by us. Deployment models → What gets deployed Koru Runtime, and the architecture diagram showing the trust boundary. The platform → Privacy policy How we handle data collected through this website. Read the policy →

A useful first conversation

Where does the work get stuck?

Bring the workflow people actually follow, including the workarounds. We will trace where it gets stuck. Then we will agree on the approvals and one result to measure after launch.

Map a workflow with our engineers

Talk directly with the engineers who build the agent.