Local-memory lifecycle for the official Hindsight plugin on DSH desktop: one-click start/stop of the hindsight-embed daemon, auto-start on boot, stop on exit, hides the uv.exe console window, and installs the vault/models to another drive.
Install
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:Hua1Q1nG/dsh-hindsight-local-optimize
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).
README
This plugin publishes its README in Chinese only.
为 DeepSeek Harness(DSH 桌面端)的 @vectorize-io/hindsight-coding-agents 插件补上「本地记忆」完整生命周期与体验的优化插件。
本插件不修改官方插件的业务逻辑,而是作为旁路补丁:接管守护进程的启动/停止、自动绑定 DSH 生命周期、隐藏终端黑框、支持把档案柜装到其它盘。
背景:解决了什么问题
官方 hindsight-coding-agents 在 DSH 桌面端有两个缺口:
本地守护进程不会自动启动 官方 DSH 版插件只把记忆地址指向本机
127.0.0.1:9077,但从不主动拉起hindsight-embed(hindsight-api)守护进程——自动冷启动只实现在 Claude Code / Cursor / Copilot 等 CLI 工具里。结果是:选「本地保存」后,DSH 去连 9077 端口但那里没人值班,记忆功能等于没生效。会弹出黑框 守护进程的宿主
uv.exe在 Windows 11 上会用WindowsTerminal弹出一个终端黑框,且常驻不消失。
本插件补齐这两点,并提供一键开关、开机自启、退出随停。
功能
- ✅ 一键开关:侧边栏「本地记忆」按钮,实时状态灯(未运行 / 启动中 / 运行中 / 失败)。
- ✅ 开机自启:DSH 启动时自动在后台拉起守护进程(带重试,约 40 秒内就绪,不阻塞 DSH)。
- ✅ 退出随停:独立「看门狗」进程监控 DSH 主进程,DSH 退出时立即杀掉守护进程、释放内存。
- ✅ 隐藏黑框:自动给官方启动器
daemon-start.js打「隐藏窗口」补丁(自愈),并后台隐藏uv.exe的终端窗口。 - ✅ 装到其它盘:守护进程(uv 缓存)与模型(HuggingFace 缓存)可安装到任意盘。
架构
``` dsh-hindsight-local-optimize/ ├── lib/ │ ├── index.js # 宿主半边:路由、启停、凭证读取、官方启动器补丁、看门狗 │ ├── client.js # 客户端半边:侧边栏开关按钮 │ ├── hide-window.ps1 # 隐藏 uv.exe 终端黑框(Win32 ShowWindow) │ └── watchdog.mjs # 看门狗:独立进程,DSH 退出时杀守护进程 ├── package.json # 双面插件声明(dsh.bundle.patch + dsh.client) ├── cordis.patch.yml # 挂载补丁 └── README.md ```
工作原理(关键点)
- 启动链路:插件读宿主凭证服务取
DEEPSEEK_API_KEY(Key 不出宿主进程)→ 注入HINDSIGHT_API_LLM_*环境变量 → 后台spawn node daemon-start.js --harness dsh拉起守护进程。 - 退出链路:因为守护进程是「孤儿进程」,且 DSH 退出时承载插件的进程会被 Windows 强杀、退出钩子不可靠,所以插件额外 spawn 一个独立看门狗进程,每 2 秒探测 DSH 主进程;一旦发现 DSH 退出,立即
taskkill守护进程并自身退出。 - 黑框处理:启动前给官方
daemon-start.js内部spawn uvx补windowsHide: true(自愈:官方更新覆盖后下次自动重打),再用hide-window.ps1枚举并隐藏标题含uv.exe的窗口。
安装
前置条件
- 已通过 DSH 插件市场安装官方
@vectorize-io/hindsight-coding-agents(本插件依赖它的daemon-start.js)。 - 本机已安装 uv(
uvx在 PATH),用于运行hindsight-embed。 - 已在 DSH 凭证里配置
DEEPSEEK_API_KEY(本地记忆抽取需要一个大模型,官方支持 DeepSeek)。
安装本插件
把本仓库克隆或复制到 DSH 的全局插件目录:
``` /profiles/node_modules/dsh-hindsight-local/ ```
在
<DSH_HOME>/profiles/web/cordis.patch.yml末尾追加一行挂载:```yaml
- insert:
- id: hindsight-local name: 'dsh-hindsight-local'
```
- insert:
完全退出并重新打开 DSH 桌面端。
使用方法
安装并重启 DSH 后:
- 自动启动:DSH 启动后约 40 秒,左下角「本地记忆」按钮会自动变成「运行中」(无需手动操作)。
- 手动开关:点击左下角「本地记忆」按钮,可随时手动启动 / 停止守护进程。
- 退出随停:直接关闭 DSH,守护进程会被看门狗自动杀掉、内存立即释放。
配置参数
插件自身的配置在 <DSH_HOME>/storages/dsh-hindsight-local/config.json:
```json { "installDir": "X:\hindsight", // 守护进程与模型的安装目录;X 代表任意盘符,留空则用系统默认位置 "autoStart": true // 是否开机自动启动 } ```
installDir:非空时,插件会把UV_CACHE_DIR指向<installDir>/uv-cache、HF_HOME指向<installDir>/hf-models,让档案柜和两个小模型装到指定盘。autoStart:true= DSH 启动时自动拉起守护进程;false= 只靠手动开关。
Hindsight 记忆本身的配置仍在 ~/.hindsight/coding-agent.json(插件启动时会自动写入 serverMode: "daemon")。
注意事项
- 资源占用:本地守护进程常驻约 0.8–1 GB 内存;首次安装约 3 GB 磁盘(主要是 torch 依赖 + 两个小模型)。关闭开关或退出 DSH 后内存即释放。
- 首次冷启动慢:第一次启动要下载
hindsight-embed包和模型,约 1–3 分钟;之后缓存命中,40 秒内就绪。 - 黑框是隐藏不是消除:
uv.exe的终端窗口被ShowWindow(SW_HIDE)隐藏,进程仍在后台;彻底不弹需等uv上游支持。别手动点黑框的 × 关闭它,那会杀掉守护进程。 - 官方更新:官方插件更新覆盖
daemon-start.js后,本插件的窗口补丁会在下次启动时自动重打。 - 密钥安全:
DEEPSEEK_API_KEY始终通过 DSH 宿主凭证服务读取,不落盘、不出宿主进程,请勿在代码或配置里硬编码密钥。
License
MIT
Links
More in this category
vectorize-io/hindsight#coding-agents★ 61444
Hindsight, agent memory that learns: long-term project memory with auto recall and retain, knowledge pages, deep reflection, and per-repo memory banks.
volcengine/OpenViking#examples/dsh-memory-plugin★ 42323
OpenViking memory and context bundle for DeepSeek Harness: pre-step auto-recall and profile injection, session capture, `viking://` URI guarding, and recall/write memory tools backed by an OpenViking server.
MemTensor/MemOS#memos-local-plugin★ 12418
Local-first memory plugin for DeepSeek Harness: layered L1/L2/L3 memory with automatic capture, manual memos_* retrieval tools, and a built-in memory viewer.
agentscope-ai/ReMe#dsh★ 3814
Connects DeepSeek Harness to ReMe's local-first, self-evolving personal knowledge base: automatically captures completed main-agent conversations as user-owned Markdown memory, searches conversations and source material through reme_search with BM25, optional embeddings, and wikilink expansion, and schedules daily memory consolidation.
zilliztech/memsearch#MemSearch★ 2892
Shared Markdown memory for DSH and other coding agents, with automatic capture, pre-step context injection, searchable recall, and memory-to-skill self-evolution through a review panel.
vshulcz/deja-vu#extensions/dsh★ 1578
Cross-session memory and context recovery across forty-five other coding agents (45+) on this machine, reading session files they already wrote before install (Claude Code, Codex, Cursor, Windsurf, Trae, Antigravity, VS Code Copilot Chat, opencode, OpenClaw, Hermes, Kimi, Qoder, Cline, Zed, Magpie and more): six tools (deja_recall, deja_session, deja_blame, deja_fix, deja_how, deja_remember), a /deja command, and optional automatic recall injected into runtime context. Local BM25 CLI/MCP index, no LLM, no embeddings, no network (dsh plugin --profile web add dsh-deja).
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.