@plur-ai/dsh
DeepSeek Harness 的 PLUR 记忆插件：将记忆单元直接注入 Prompt 中，无需调用工具。
AI 分析
核心用途是为 DSH 智能体提供免工具调用的记忆注入功能。适合希望智能体能自动检测并记住用户纠错、跨会话保持上下文记忆的 DSH 用户。
安裝
$
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:plur-ai/dsh-plugin說明文件
閱讀完整 README ↗Configuration
All settings live under the plur namespace in $DSH_HOME/settings.yaml
(usually ~/.dsh/settings.yaml).
| Setting | Default | Meaning |
|---|---|---|
path | ~/.plur | Store location |
scope | derived | Which memory scope this harness may read and write. Omitted, it derives per workspace |
injectionMode | content | content injects memories; off disables injection |
injectionBudget | 2000 | Approximate token ceiling for the injected block |
refreshIntervalMs | 0 | Floor between recalls; 0 means once per turn |
autoLearn | true | Detect corrections in your messages and store them |
autoCapture | true | Record an episode summary at turn end |
timeoutMs | 5000 | Hard bound on any single memory call |
viewerEnabled | true | Register the /plur-memory command |
includeGlobal | true | Whether global engrams accompany the workspace scope |
Reranking is configured through PLUR itself, not here: set the
PLUR_RERANKER environment variable. It runs in the harness's own process and
bge-reranker-v2-m3 peaks around 2GB RSS, where a native OOM cannot be caught
by a JavaScript try/catch and would take your agent down with it. Leave it
unset for interactive use; enable it for local batch work where a crash
costs you nothing.