Skip to content
aicerineDocs

Aicerine documentation

Aicerine keeps your OpenStack hosts at the overcommit ratio you chose without letting them run hot. It does it with two levers, both through public OpenStack APIs, with nothing installed in Nova:

  1. Prevention. Aicerine sets the VCPU allocation ratio of each host in Placement. A host that runs hot stops admitting new VMs; it admits them again when it cools down.
  2. Correction. Aicerine plans live migrations with an explicit cost, replays each plan in a simulator, then runs it at the autonomy level you set. It often moves the neighbours of a large VM rather than the large VM itself, because that is cheaper.

Aicerine runs as an appliance on your infrastructure, air-gapped if needed. A read-only agent on each compute host measures what VMs actually consume.

What holds everywhere

  • No fork, no patch of Nova. Nova keeps its filters, its weighers and the final selection. Placement stays the source of truth.
  • Decisions are deterministic. The same state and policy give the same plan, and every move carries a calculation an operator can redo by hand.
  • You choose who acts. Aicerine starts in observe and only migrates when you allow it. An emergency stop freezes everything.
  • The LLM explains; it never decides. It is off by default, and the product works without it.
  • Nothing leaves your infrastructure by default.

Where to start