WhaleHarness/WhaleHarness--plugins-whale-memory ↗★ 0
whale-memory
Explicit cross-session memory for DSH: the model can remember and recall across conversations. Pure local file storage, no subprocess, no network.
AI 분석
核心用途是实现安全、可控的跨会话长期记忆。适合需要 Agent 在不同对话间记住特定事实,且注重隐私安全的用户。
설치
$
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:WhaleHarness/WhaleHarness#d3888b29bb0991e2b08a8099683e87134bc37680&path:plugins/whale-memorywhale-memory
Explicit cross-session memory for DeepSeek Harness. The model decides what is worth remembering, tells the whale, and any later session on the same machine can recall it.
Install
dsh plugin --profile web add -w https://whaleharness.com/plugins/whale-memory-0.1.0.tgz?src=install
Tools
whale_memory_set(key, value)— remember one named fact (overwrites)whale_memory_get(key)— recall itwhale_memory_list()— every key with last-updated time, newest firstwhale_memory_delete(key)— forget it
Storage & safety
- One JSON file:
~/.dsh/storages/whale-memory/memory.json(atomic replace writes) - No subprocess, no network, no credential access — the WhaleHarness red lines
- Limits: 1000 keys, 64KB per value, key charset
[A-Za-z0-9._:-]{1,128}
How it differs from the built-in session reference
DSH's built-in cross-session reference is passive and read-only (it surfaces candidate snapshots). whale-memory is explicit and writable: the model curates what is remembered, so continuity becomes a decision, not an accident.