Actor Ownership Moves to a Dedicated Database Column
A technical engineering account describes how an AI memory system evolved from storing actor ownership in JSON metadata, through namespace-prefix encoding, to a dedicated column in its memories table. The column enables exact SQL matching, while ActorVisible centralizes scope-dependent read visibility and RequireActor validates writes within saveCore. The report does not identify the memory engine, implementation repository, or deployment context.
The case illustrates how a dedicated ownership column, fail-closed visibility rules, and a centralized write guard can support precise filtering while keeping namespaces focused on topics.
2026-10-08·记忆系统·Actor 边界
引擎的 actor 归属经历了三个阶段:metadata JSON 键 → namespace 前缀编码 → 独立数据库列。每个阶段看起来都是"正确的修法",但每个阶段都引入了新的一类 bug。这是一个关于"用什么数据结构表达归属"的工程故事。
一、阶段 1:metadata JSON 键
最初的设计很自然——归属信息放在 metadata 里:
{
"content": "我喜欢简洁风格",
"scope": "global",
"metadata": {
"actor": "alice",
"source": "conversation"
}
}
#AI memory system#Mem0#LangGraph#AWS AgentCore#MUXI#Oracle AI Memory#Agentium#Letta