dsh-memory-evolve
其他 活跃维护

dsh-memory-evolve

brianwang2007xjtu/dsh-memory-evolve

提供跨会话长期记忆能力,支持待办事项、个人技能的统一管理,可编排多个AI会话及外部智能体协同工作,随使用自动迭代适配需求。

1
Stars 标星
0
Forks 分支
1
Watchers 关注
0
Open Issues
JavaScript
主要语言
MIT
开源协议
6 KB
仓库大小
1 个月前
最后推送
一键安装扩展 / 插件指令
dsh plugin --profile web add github:brianwang2007xjtu/dsh-memory-evolve
git clone https://github.com/brianwang2007xjtu/dsh-memory-evolve.git
git clone git@github.com:brianwang2007xjtu/dsh-memory-evolve.git
README.md main

dsh-memory-evolve

Evolving long-term memory for the DeepSeek Harness: durable, decay-reinforced
memories persisted through the storage-domain form (JSON backend under
$DSH_HOME/storages).

Tools

  • memory_remember — add/update a memory; near-duplicate content merges
    into the existing entry instead of creating a duplicate.
  • memory_recall — retrieve memories by query/tags, ranked by relevance +
    importance + recency; reading a memory reinforces it.
  • memory_evolve — force one consolidation pass: decay importance by age,
    merge near-duplicates, prune below the retention threshold.
  • memory_forget — delete one memory by id, or clear the whole workspace.

Memories are namespaced by the session's workspace (cwd), falling back to
namespace for sessions without a workspace.

Configuration

- id: memory-evolve
  name: dsh-memory-evolve
  config:
    namespace: global      # fallback namespace
    maxMemories: 1000      # per-namespace cap (oldest pruned first)
    decayHalfLifeDays: 30  # effectiveImportance = importance * 0.5^(age/halfLife)
    pruneBelow: 0.05       # prune memories whose effective importance drops below this
    mergeThreshold: 0.8    # Jaccard similarity above which memories merge
    recallLimit: 10        # default recall limit

Installation

dsh plugin --profile web add link:/path/to/dsh-memory-evolve

Then mount it with the patch row above in the profile's cordis.patch.yml.