MIT 许可的本地优先 agent memory 中枢 + 本地 Agent 运行时：通过 `memmy-memory init` 安装的 memory Skills 与本地记忆 API（127.0.0.1:18960），让 DeepSeek Harness 类编码 Agent 共享同一套长期记忆，支持桌面应用、CLI/TUI 与 BYOK。
DSH 适配
兼容
作者声明
安全审计
未审计
最后核验
2026-08-21
许可证
MIT
01它能帮你完成什么？
Give every coding agent one shared, persistent agent memory so project context carries over between agents and sessions
A local-first memory layer that keeps project goals, decisions, constraints, and failed attempts, then brings the relevant context to the next Agent — no need to re-introduce anything again
Developers running DeepSeek Harness-style coding agents (Codex, Claude Code, Cursor, OpenClaw) whose agents keep losing context between sessions
Convert months of existing agent history into searchable long-term memory within minutes
Automatic scan of Cursor, Claude Code, Codex, OpenCode, OpenClaw, Hermes, WorkBuddy, Pi and qwenwork histories, distilled into personal long-term memory plus a personalized "First Meeting Report"
Teams and individuals with months of conversations in other agents who want that context to become a reusable knowledge asset instead of starting from scratch
Wire memory search and writes into external agents and scripts
`memmy-memory` CLI (init / health / search / add / get) against the local memory service at http://127.0.0.1:18960, with memory Skills installed into each supported agent as needed
Developers who want agents or shell scripts to query and append shared memory programmatically, including coding agents like DeepSeek Harness via installed Skills
02如何接入 DeepSeek Harness？
前置条件
- Memmy desktop app or CLIs, downloaded from the official website (https://memmy.bot/) or GitHub Releases
- An account (sign-up grants Agent task trial tokens) or BYOK: your own model API key configured in ~/.memmy/config.yaml
- Only for building from source: Node.js >= 22 and npm
安装步骤
- 01
Get Memmy from the official website (https://memmy.bot/) or GitHub Releases, launch the desktop app and choose Account mode or API Key mode
- 02
In API Key mode, configure the primary model and pass a connection test; optionally configure Embedding, ASR, image generation, memory summary, and skill evolution models
- 03
Enter the main workbench and send your first task; open "Tools" to connect messaging channels or third-party tools, open "Memory" to scan Agent history sources
- 04
To give external agents memory access, run `memmy-memory init` — it writes the Memory config and installs Skills for each Agent as needed
验证接入成功
- Run `memmy status` to check config, workspace, model, and provider status
- Run `memmy-memory health` to check the memory service
03DSH 适配与能力边界
Local agent memory hub whose `memmy-memory init` CLI writes the Memory config and installs memory Skills into external agents, letting DeepSeek Harness-style coding agents read/write one shared long-term memory via the local memory service; the repo is tagged `dsh-plugin`, though the README documents no dsh-specific flow
Cross-Agent shared long-term memory
project goals, decisions, constraints, failed attempts and agent conversations→one shared memory layer reused across Codex, Claude Code, Cursor and OpenClaw without re-introducing context
memory, configuration and app state are written to your machine by default (~/.memmy)history onboarding reads the local history files of your existing agentsMemOS-powered memory engine with history onboarding
scattered conversations and behavior from supported agents (Cursor, Claude Code, Codex, OpenCode, OpenClaw, Hermes, WorkBuddy, Pi, qwenwork)→structured, searchable, reusable long-term memory plus a personalized "First Meeting Report"
scanning imports and stores other agents' history into Memmy's local memoryLocal Agent runtime with multiple entry points
tasks issued from the desktop app, `memmy` CLI/TUI, or any OpenAI-compatible client→the same Agents, memory, and configuration served from every entry point; `memmy serve` exposes an OpenAI-compatible API on :18990
`memmy serve` opens a local API on port 18990; the memory service listens on 127.0.0.1:18960Extensible tools via Skills and MCP
Skills, MCP servers, messaging channels (Telegram, Discord, WeChat, Feishu, DingTalk) and productivity tools (GitHub, Gmail, Notion, Slack, Jira)→agents that go from conversation to real task execution, including managed Chromium browser tools for local page inspection
the desktop app and scripts/dev-start.sh prepare a managed Chromium build before the Agent Gateway startsconnected third-party tools and channels are invoked over the network under your credentials
04适合谁？何时不该用？
适合
- Developers running DeepSeek Harness-style coding agents (Codex, Claude Code, Cursor, OpenClaw) whose agents keep losing context between sessions
- Teams and individuals with months of conversations in other agents who want that context to become a reusable knowledge asset instead of starting from scratch
- Developers who want agents or shell scripts to query and append shared memory programmatically, including coding agents like DeepSeek Harness via installed Skills
05兼容性、维护与安全提示
- Account mode grants Agent task trial tokens whose amount and usage are shown in the app; once the trial credits run out or expire, continued agent tasks require switching to BYOK with your own model API key.
- Building from source requires Node.js >= 22 and npm, and on Windows scripts/dev-start.sh must be run in Git Bash.
- The README documents memory-Skill installation for external agents generically and the repo is tagged `dsh-plugin`, but no DeepSeek Harness-specific setup is documented — usage inside dsh is inferred via the installed memory Skills / MCP ecosystem.
MIT · actively maintained (latest release v1.0.9, 2026-08-20)
06常见问题
Memmy 如何接入 DeepSeek Harness 这类编码 Agent？
运行 `memmy-memory init`，它会写入 Memory 配置并按需为各个 Agent 安装 memory Skills，之后这些 Agent 通过本地记忆服务（默认 http://127.0.0.1:18960）读写同一套记忆。README 未提供 dsh 专属步骤——仓库标注了 `dsh-plugin` 主题，接入走的是通用的 Skills / MCP 生态。
记忆数据存在哪里？会上传云端吗？
Memmy 采用本地优先架构：记忆、配置和应用状态默认保存在你的机器上，无需上传云端。记忆服务监听 127.0.0.1:18960，并提供受控访问机制，只有授权来源才能调用记忆能力。
需要自己准备 API Key 吗？
账号模式注册即送 Agent 任务体验额度（当前额度与用量以应用内显示为准）。额度用完或过期后，切换到 API Key（BYOK）模式，在 ~/.memmy/config.yaml 中配置自己的模型服务商即可。
支持导入哪些 Agent 的历史记录？
目前支持 Cursor、Claude Code、Codex、OpenCode、OpenClaw、Hermes、WorkBuddy、Pi 和 qwenwork。几分钟内即可把历史对话转换为长期记忆，并生成个性化的「初次见面报告」。
Memmy 有哪些使用入口？
桌面应用、`memmy` CLI/TUI，以及通过 `memmy serve` 启动的 OpenAI 兼容 API（端口 :18990）——三个入口共享同一套 Agent、记忆与配置。
07相关的 DSH 工作流
ruflo
作者 ruvnet
原版智能体元框架（meta-harness）：支持多智能体集群部署、自主工作流编排与对话式 AI 系统构建，内置自适应记忆、自学习智能与 RAG 检索，原生集成 Claude Code / Codex / Hermes 等大量工具。
reactive-resume
作者 amruthpillai
注重隐私的开源简历生成器，安全、可定制、可移植，永久免费，支持自托管。
everos
作者 evermind-ai
为所有 AI 智能体提供的便携统一记忆层，本地优先、Markdown 原生、数据归用户所有，跨应用与工作流自进化。
yao
作者 yaoapp
在桌面、移动端、浏览器或 API 上统一管理你的智能体与工作区，看板式任务追踪，支持自托管。
08数据与来源
memmy-memory init # Write the Memory config and install Skills for each Agent as needed
Connect more tools through Skills and MCP, taking the Agent from conversation to real task execution.
页面基于项目公开文档、仓库元数据和 DSH Plugins 的结构化解析生成；最后核验于 2026-08-21。发现错误？提交更正。
最佳 DeepSeek Harness 插件
从全目录挑出的 12 个值得优先安装的插件，覆盖各个分类。