Security and governance

Agents that work inside your operations, by your rules

Your data stays yours. Every action an agent takes is recorded and traceable, and a person on your team approves it before anything is written to your systems.

Commitments

Four commitments in every deployment

Close-up of a processor on a circuit board
  1. 01

    Zero data retention

    Your data never trains a model, ours or anyone else's. The model providers a deployment calls work under zero-retention agreements: they keep nothing once a request is answered.

  2. 02

    Deployment and isolation

    Mirage runs where your policies require, from the cloud to a server on your premises. Each deployment is isolated: its own environment and its own data, shared with no other client.

    • Cloud
    • Private cloud
    • Your cloud account
    • Sovereign cloud
    • On premises
    • Edge and offline
  3. 03

    Full audit trail

    Every agent action is logged with the inputs it received, the sources it relied on and the result it produced, so any figure can be traced back to the document it came from.

  4. 04

    Access and human approval

    Access rights are set by role. Before an agent writes anything into your systems, a named person on your side approves it.

How control works

From an incoming document to a validated entry

The agent prepares, a person decides, and the log keeps both. Here is the path a subcontractor invoice takes before anything reaches the ERP.

See approvals and audit
Lines of code on a screen, out of focus
How control worksIllustrative example
  1. Input

    A subcontractor invoice arrives by email with 14 line items.

  2. Agent action

    The agent checks each line against the purchase order and the signed delivery notes, and flags one line billed above the agreed rate.

  3. System action

    The ERP entry is prepared for the 13 matching lines. The flagged line is held, with its sources attached.

  4. Human decision

    The site manager reviews the flagged line, rejects it and approves the other 13.

  5. Outcome

    The approved lines are written to the ERP. The action, its sources and the approver's name go into the audit trail.

Questions your IT team will ask

  • Your data never trains a model, ours or anyone else's. The model providers a deployment calls work under zero-retention agreements and keep nothing once a request is answered. When Mirage runs in your infrastructure or your cloud account, data and processing stay in the environment you control.

  • In the cloud, a private cloud, your own cloud account, a sovereign cloud, on your premises, or at the edge and offline. Each deployment is isolated: its own environment and its own data, shared with no other client.

  • Through the interface each vendor provides: an API, a file exchange, eAdaptor for CargoWise, data entities over OData for Dynamics 365. Where a system has no API, Mirage builds the connector during the deployment. Nothing is migrated. See the integrations

  • Mirage's access is limited to what the deployment needs, and you can remove it. In a deployment in your infrastructure or your cloud, the encryption keys are yours.

  • Access rights are set by role. Approvals and the thresholds that send an item to a person are set for each workflow, and every agent action is logged with its inputs, sources and result. See approvals and audit

  • Yes. Send it to mehdi@miragemetrics.com.

  • Yes. The AI policy guide gives contractors and joint ventures a free policy template (.docx), the clauses an owner can put in an RFP, and a vendor evaluation checklist. It is a starting point for your own policy, not legal advice. See the AI policy guide

Next step

Automate the coordination layer of complex operations

Bring one workflow. We will show you where the agent acts, where your team approves, and what the log keeps.

Get in touch

Security questions or a questionnaire to complete: mehdi@miragemetrics.com