Infrastructure experiments for governed agent execution.

AIRBQ explores one narrow question: how can an agent execute controlled workflows with explicit checkpoints, compensation paths and evidence a human can inspect?

Concept / Incubation — no hosted runtime, commercial API or hardware deployment.
Run the browser simulation Read the current boundary

Three questions worth validating

The project stays in incubation until these mechanisms can be demonstrated beyond a concept page.

01 / Control

Can execution be bounded?

Actions should begin from an explicit grant, remain inside a declared scope and stop at a human gate when consequences expand.

02 / Recovery

Can partial work be compensated?

A failed multi-step workflow needs a visible recovery path, not an optimistic claim that every action is reversible.

03 / Evidence

Can the result be inspected?

Operators need an event trail that distinguishes proposal, execution, failure, compensation and final state.

SAGA concept demo

This is a deterministic browser simulation. It does not contact infrastructure, an agent, a model or a physical device.

simulation: true
Choose a pre-scripted outcome to inspect the proposed state model.
SAGA observationidle
{
  "simulation": true,
  "state": "awaiting_scenario",
  "external_effect": false
}

Current boundary

AIRBQ is a low-power infrastructure research slot, not a deployed execution network.

  • No public account, hosted runtime or commercial endpoint.
  • No claim of production reliability or physical-world deployment.
  • No shared product matrix: related work is curated by IPIPQ.
  • Expansion requires a real user task and reproducible execution evidence.