Security and data sovereignty

How NomOS controls access and data flows

NomOS governs which context a model receives, which action an agent may execute and which evidence is created. How far that control extends depends on the operating model and technical integration. The following sections explain these controls and their limits.

Four decisions define the security model.

Product logic remains the same in SaaS and Self-Hosted. Differences arise in isolation, model operation and operating responsibility. The complete choice is set out in the offering.

  • Where does NomOS run?

    As SaaS in the agreed cloud environment, or as Self-Hosted on your Kubernetes.

  • Where does the model run?

    Both operating models can connect compatible external or locally hosted models. The selected endpoint determines where approved context is processed.

  • Who receives access?

    Rooms, identity and roles determine what people and agents may retrieve before access occurs.

  • What remains traceable?

    The source used, accountable owner, applied policy, approval and governed action remain connected.

Controls apply before, during and after AI use.

  1. Before the request

    NomOS checks the user, client or agent identity, room and access rights. Configurable checks for personal data and secrets run before content is passed on.

  2. Before the action

    Policies evaluate planned agent actions. The result is to allow, escalate for human approval or block. This statement only applies to technically connected paths.

  3. After the use

    Governed answers and runs leave structured evidence. Depending on the agreed integration, events can also be sent to the organisation's own SIEM, including QRadar or Splunk. We agree event coverage, signing, separate key management and retention for your installation. A copy of the evidence in your SIEM makes later changes visible as well.

Limits to consider in your security assessment

  • Control applies to connected paths.

    Anything bypassing NomOS is neither checked nor evidenced. Each installation therefore declares its actual enforcement level.

  • Check access outside NomOS separately

    An evidence record shows what happened on a controlled path. Additional infrastructure and endpoint controls are needed to assess activity outside that path.

  • What content scans detect

    Detection depends on configured entities and thresholds. Content outside that configuration is not automatically detected.

  • Your organisation assesses legal requirements

    Legal classification, risk management and the content of the policies remain the organisation's responsibility.

These details are agreed before operation begins.

  • Region and data location

    We record the agreed region in the technical and contractual documentation.

  • Subprocessors

    We name all providers used for SaaS in the due diligence and contractual documentation.

  • Support access

    Roles, approval path, time limit and logging are defined for the specific installation.

  • Certifications and external audits

    This page currently makes no certification claim. Completed reviews are only named once reliable evidence exists.

Report a vulnerability

Report vulnerabilities to hello@ainomos.ch. We confirm receipt within one working day. Please do not include credentials or personal data.

Distinguish MCP access from local execution

NomOS checks identity and permissions for access to its capabilities. Agents can also request an action assessment. That response alone cannot force an external tool to follow it. Technical blocking requires an execution path that enforces the decision. NomGate is planned to add local control.

MCP capabilities and control scope ↗