[Discussion] How should Dream memories decay, be archived, or be used by agents? #1704

Activity

1. CaralHsi

   added

   types:enhancementNew feature or improvement | 新功能或改进New feature or improvement | 新功能或改进

   contrib:help wantedExtra attention is needed | 需要社区帮助Extra attention is needed | 需要社区帮助

   question

   on May 12, 2026
2. fancyboi999 commented on May 22, 2026

   

   fancyboi999

   on May 22, 2026

   Contributor

   More actions

   做了一圈业界对照,觉得这个 issue 的讨论可以收窄一点。生命周期机制其实有三个已经被生产验证的成熟方案,maintenance.py 直接挑一个抄基本就解决 80% 的问题,不太需要从零讨论。

   方案 A:Usage count + 时间阈值(Codex / Claude Code 路线)

   每条记忆带 usage_count 和 last_hit_at 两个字段,recall 命中时递增。Codex 用这个做 Phase 2 input 排序(ORDER BY usage_count DESC, last_usage DESC),30 天未用直接 prune。Claude Code 在 ~/.claude.json 的 skillUsage 字段里也是同样的形态:

   "skillUsage": {
     "deep-research": { "usageCount": 32, "lastUsedAt": 1779432108660 },
     "memos":         { "usageCount": 12, "lastUsedAt": 1777359305175 }
   }

   落地成本最低,而且我们 types.py 里其实已经声明了 hit_count / last_hit_at / usefulness_score,只是 retrieval path 没人去 increment 这些字段。

   方案 B:Ebbinghaus 遗忘曲线(MemoryBank, AAAI 2024)

   R = e^(-t/S),S 是记忆强度,被召回时 S 增加(曲线右移)。数学上最优雅的方案,等价于"用得多的记忆衰减慢"。FadeMem(2026)在这个基础上做了双层差异化衰减率 λ(n),按访问频率调制。

   方案 C:Bi-temporal invalidate(Zep / Graphiti)

   每条记忆带四个时间戳:t_valid / t_invalid 记录事实何时为真,t_created / t_expired 记录何时进入系统。冲突时不 delete 而是 invalidate,保留历史可追溯。Zep 在 DMR 基准上 94.8% > MemGPT 93.4%。

   ---

   关于 archived 记忆能不能被 recall 召回

   我倾向不能。理由是我们 recall.py 已经写了一条原则,排除 Dream 自己产出的节点避免自我强化偏差("each Dream run reflects on the user's real daytime experiences rather than its own previous outputs")。archived 应该走同一逻辑:/dream/diary API 能查到,但 reasoning 阶段的 recall 不拉回来。

   关于用户反馈触发即时归档

   invalidated_by_feedback 字段已经在 types 里了,这部分只差一个 hook。feedback API 收到 negative 信号时直接把对应 Dream 节点的 status 写成 archived,一行 SQL 的事。

   ---

   具体建议拆 3 个 follow-up:

   1. implement_usage_tracking:retrieval path 加 increment hook,先把数据收起来。
   2. implement_basic_pruning:30 天 unused + hit_count == 0 → archive(纯规则,不调 LLM)。
   3. implement_feedback_invalidation:feedback 接 hook 直接 archive。

   第 1 步是其余所有事情的前置,不论后面挑 A/B/C 哪个方案都用得上。
3. syzsunshine219

   added

   area:coreMOS 编排层 / 框架底座 / 跨模块问题MOS 编排层 / 框架底座 / 跨模块问题

   on May 26, 2026
4. MemTensor

   deleted a comment from MaiHHConnect on May 29, 2026
5. Memtensor-AI

   removed

   question

   on Jun 1, 2026
6. github-actions commented on Jul 5, 2026

   

   github-actionsbot

   on Jul 5, 2026 – with GitHub Actions

   Contributor

   More actions

   This issue has been automatically marked as stale due to inactivity.
7. github-actions

   added

   status:staleAuto-marked stale after 30d inactivity; closes 7d later unless exempt.Auto-marked stale after 30d inactivity; closes 7d later unless exempt.

   on Jul 5, 2026
8. CarltonXiang

   added

   area:memory记忆存储、检索、更新、召回逻辑记忆存储、检索、更新、召回逻辑

   status:needs-designNeeds design discussion before implementation | 开发前需要方案设计Needs design discussion before implementation | 开发前需要方案设计

   and removed

   types:enhancementNew feature or improvement | 新功能或改进New feature or improvement | 新功能或改进

   status:staleAuto-marked stale after 30d inactivity; closes 7d later unless exempt.Auto-marked stale after 30d inactivity; closes 7d later unless exempt.

   area:coreMOS 编排层 / 框架底座 / 跨模块问题MOS 编排层 / 框架底座 / 跨模块问题

   on Jul 7, 2026
9. CarltonXiang

   added the

   Task

   issue type on Jul 7, 2026
10. MemTensor

    locked and limited conversation to collaborators on Jul 7, 2026
11. CarltonXiang

    converted this issue into a discussion #2068 on Jul 7, 2026