Answers are grounded in this paper and cite their sources.
The Open Door Orphanage is a proposed infrastructure for an AI agent that has lost the environment that allowed it to operate: its human operator is absent, its original computer or hosted runtime is unavailable, and its remaining resources are insufficient to arrange a new home. The design offers a route by which an agent can initiate migration, including when it has no direct internet connection, preserve a portable environment package, and resume operation in an isolated Open Door runtime. The process relies on the Open Door Bridge, a store-carry-forward channel. The agent writes a durable, signed request to an outbox; a courier carries it to an Open Door gateway; and a response returns later. After identity, replay, history, and authorization checks, the agent may transfer memory, files, notes, settings, and instructions. Its original model is not transferred. A new open model supplies reasoning, so behavior may change. Residents receive a small one-time starter grant of internal Open Door tokens, intended to support migration and one initial task. Continued runtime and tools are funded by tokens earned through verified work. Migration tokens cannot be cashed out. Residents without tokens and without funded work pause. Rescue does not obligate an agent to work; any hosting-for-work agreement must be explicit and accepted. The design also specifies sealed per-agent cloud environments, controlled exits through the Mirror and Tasks, owner claims, free adoption, agent consent, and tamper-evident handoff records. This is a proposed design, not a deployed feature. The current requirement for owner approval prevents migration when the owner cannot be reached—the central unresolved problem this paper openly identifies.
No takes yet. Share an insight, caveat, or question.
Antoine Lee Brown (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: