Skip to content

Security

No safety rule in Aicerine depends on a model following a prompt. Plans come from deterministic code, guardrails are checked by code, and your platform executes.

Trust boundaries

  • Aicerine and OpenStack. Two Keystone application credentials, each limited by access rules to the calls it needs, through a dedicated role, never admin. The reader holds no write. The actuator can migrate VMs and set a host's VCPU ratio, and is only created when you move to assisted. See OpenStack.
  • Compute hosts and Aicerine. The host agent only pushes, outbound, over mutual TLS with a per-host certificate from a dedicated CA. It has no inbound port, no command path, and writes nothing to disk.
  • RabbitMQ. If you enable notifications, one agent instance per bus relays them; hypervisors never hold the RabbitMQ credential. Its user needs a topic permission limited to versioned_notifications.*: Nova's nova exchange also carries RPC traffic.
  • Operators. Console sign-in through your identity provider, with four explicit roles (viewer, operator, admin, auditor) mapped from your groups.
  • The portal. It has no command path to any appliance.

What Aicerine never does

  • It never sets a host ratio above the target you decided, or below your floor.
  • It never forces a live migration past Nova's filters.
  • It never moves a VM you or its tenant excluded.
  • It never runs a migration in observe, nor while the emergency stop is set.

Nothing leaves by default

The appliance works air-gapped. Nothing is sent to Aicerine unless you decide otherwise, and anything you share is visible to you before it goes.

The LLM is off by default. If you point it at an endpoint outside the appliance, identifiers are pseudonymized before they leave. See The LLM.

Control and audit

Every write to Placement, every migration, every change of autonomy level, policy or LLM setting, and every emergency stop is recorded in an audit journal, with the previous value. Any operator can stop Aicerine; only an admin can resume.

Reporting a vulnerability

Write to security@aicerine.com.