Security

Carina runs tasks. Scout controls risk.

This page is for teams shipping agents to production. Carina runs the work; Labyrinth Scout adds monitoring, response controls, and reporting when configured on labyrinthscout.com.

Who it is for

Platform and security leads running Carina (or another agent) where tool calls need oversight and a fast stop button.

Problem it solves

Agents can call shell, HTTP, and email tools. You need a record of what ran and a way to halt a bad session without SSHing into the host.

Outputs you get

  • Carina: task results, files, and normal runtime output
  • Scout (when paired): live monitoring, response controls, and audit reporting
  • Public read: risk metrics at carinaai.uk/trust from Scout telemetry

Config and integrations

  • Carina: pair with Scout during setup using your provisioning details
  • Scout: production monitoring, response controls, and reporting for connected agents
  • Optional trust and attestation features appear only when the operator config is active

Example workflow

  1. Carina completes a file-and-research task for an internal user.
  2. Scout records the session and raises an alert when something looks unsafe.
  3. The operator reviews the event and applies a response action from the Scout console.

Pair Scout during setup, then run normal Carina tasks; review events on your provisioned Scout console.

Labyrinth Scout Integration docs

For deeper implementation details, use the private operator documentation. The public page only describes outcomes and response options.