MindOS — Where You Think, Agents Act

01. THE VISION

人机共享心智

Human-AI Shared Mind

你在思考中沉淀认知，Agent 在执行中放大行动。人和 AI 不再一次次从零开始，而是在同一套可治理的上下文中共同成长。

You compound cognition through thinking, and Agents amplify action through execution. Human and AI grow together in one governed context instead of starting from zero every time.

全局同步 — 打破记忆割裂

Global Mind Sync — Breaking Memory Silos

换个工具、开个新对话，就要把背景重讲一遍？

Switch tools or start a new chat, and you have to re-explain everything?

01

痛点Pain

切换工具带来上下文割裂，知识无法跨 Agent 复用
个人深度背景散落各处，Agent 缺乏全局 Context

Switch tools or start a new chat and you're re-transporting context — knowledge can't be reused across Agents
Deep personal context scattered everywhere — Agents always lack the full picture

跃迁Shift

内置 MCP Server，所有 Agent 零配置直连核心知识库
项目记忆与 SOP 仅需一处记录，即可全量赋能所有 AI 工具

Built-in MCP Server — full-lineup Agents connect with zero config
Profile, SOPs & project memory: record once, empower all Agents

透明可控 — 消除记忆黑箱

Transparent & Controllable — No Black Boxes

Agent 在记忆你，但记了什么、记对没有？

Your Agent is memorizing you — but what exactly did it remember, and is it right?

02

痛点Pain

Agent 记忆锁在黑箱中，推理无法审查，错误极难追溯纠正
交互幻觉不受控地累积，人机信任链随时间持续崩塌

Agent memory locked in black boxes — reasoning can't be audited, errors can't be traced or corrected
Hallucinations compound unchecked, eroding trust over time

跃迁Shift

检索与执行均沉淀为本地纯文本，提供完整的审查干预界面
人类拥有绝对的心智纠偏权，可随时校准 Agent 行为

Every retrieval, reflection & action saved as local plain text, with a full GUI for audit and intervention
Humans hold absolute mind-correction rights — recalibrate Agents anytime

共生演进 — 经验回流为指令

Symbiotic Evolution — Experience Flows Back as Instructions

你和 AI 对话了 100 次，这些思考去了哪里？

You've thought with AI 100 times — where did all that thinking go?

03

痛点Pain

反复表达偏好，新对话又从零开始，思考未成 AI 能力
AI 不成长你也没变清晰，思考过程未沉淀为方法论

You keep expressing preferences, standards, and judgment to AI, but the next chat starts from zero — your thinking never becomes AI's capability
AI doesn't grow, and neither do you — the thinking process never crystallizes into your own methodology

跃迁Shift

思考自动沉淀为知识库，下次应对更默契，拒绝重复犯错
在交互中厘清想法与标准，认知随每次沉淀变得锐利

Every thought auto-distills into your knowledge base — AI understands you better next time, never repeating the same mistake
The act of thinking forces you to clarify your own standards and boundaries, sharpening your cognition with each iteration

底层基石：本地优先

Foundational Pillar: Local-first

所有数据以纯文本形式存储在本地，彻底消除隐私顾虑，确保你拥有绝对的数据主权与极致的读写性能。

All data is stored locally as plain text, eliminating privacy concerns and ensuring absolute data sovereignty with ultimate read/write performance.

极速响应 · 极简界面 · 键盘优先 · 本地优先

Speed First · Minimal Chrome · Keyboard-driven · Local-first

02. THE FLOW

想法、执行到复盘，一条线贯穿

From Ideas to Execution to Review, One Thread Through All

记一句想法，剩下的全自动

Jot down one idea, everything else is automatic

1 随手记录Capture 手机 / 任意设备Phone / Any Device

9:41MindOS●●●

‹快速记录Quick Capture

今天 11:30

Today 11:30

想好了新项目：先理清思路，代码开搞，顺便发个帖宣传下

New project idea: plan the approach first, start coding, post about it to get feedback

#idea

昨天 16:00

Yesterday 16:00

之前调研过竞品，用 React + Tailwind 方案

Researched competitors earlier, going with React + Tailwind stack

#tech-stack

记一下新项目的想法...Jot down a new project idea... ↑

2 自动整理Auto-Organize MindOS GUI

MindOS — localhost:3000

Projects/新项目计划.md

Agent 已自动更新 3 个文件Agent auto-updated 3 files

新项目计划

New Project Plan

核心思路

Key Approach

先理清项目方案和架构
Plan the project approach and architecture
搭建代码骨架，快速出原型
Scaffold codebase, ship a quick prototype
发帖宣传，收集早期反馈
Post for promotion, collect early feedback

关联更新

Related Updates

Projects/Project-Plan.md Workflows/Launch-SOP.md TODO.md

3 所有 Agent 执行All Agents Execute via MCP

ALL AGENTS via MCP

Gemini

› 梳理项目方案和架构Plan project approach and architecture

Reading MindOS → Project-Plan.md

Done. 已整理项目方案 + 技术架构文档 Done. Created project plan + architecture doc

Cursor

› 搭建项目骨架Scaffold project structure

Reading MindOS → Project-Plan.md

Done. 已初始化项目结构 + 基础配置 Done. Initialized project + base config

O OpenClaw BOT

发帖宣传项目，收集大家反馈

Post about the project, collect feedback

Writing → Workflows/Launch-SOP.md
Done. 已发布宣传帖 + 汇总反馈沉淀到 SOPDone. Published post + distilled feedback to SOP

一句想法 → MindOS 归档 + 关联 → 所有 Agent 各就各位 → 经验自动沉淀

One idea → MindOS archives + links → All Agents mobilize → Experience auto-distilled

03. THE DIFFERENCE

同一个任务，两种体验

Same Task, Two Realities

选择一个真实场景，看看有 MindOS 和没有 MindOS 的区别。

Pick a real scenario. See what changes when your Agents share your mind.

✕ 没有 MindOSWithout MindOS

周一 9:00，市场负责人说："下午 3 点前把这 30 位 KOL 的外联初稿发我"

Monday 9:00 AM: "Send first-draft outreach for 30 influencers before 3 PM."

1你把表格、历史合作记录、备注一条条复制进 PromptYou manually paste sheets, collaboration history, and notes into the prompt
2Agent 先给通用模板，你再逐个补充语气、内容方向和禁用词Agent returns generic templates; you rewrite tone, content angle, and blocked terms one by one
3第 19 封才发现用错人设，整批重改You catch a wrong persona at email #19 and rework the whole batch

~45 min~45 min 重复喂上下文，返工风险高context repetition + high rework risk

✓ 使用 MindOSWith MindOS

同一句话："下午 3 点前，把这 30 位 KOL 外联初稿发我"

Same sentence: "Draft outreach for these 30 influencers before 3 PM."

1Agent 自动读取 Resources/Influencers.csv + 合作历史标签Agent auto-loads Resources/Influencers.csv and collaboration-history tags
2按 Workflows/Outreach-SOP.md 生成分层外联：头部 / 腰部 / 长尾Follows Workflows/Outreach-SOP.md to generate tiered outreach: top/mid/long-tail
3一次产出可直接发送版本，你只做最终确认Produces send-ready drafts in one shot; you only do final approval

~6 min~6 min 流程自动对齐，几乎零返工workflow auto-aligned, near-zero rework

✕ 没有 MindOSWithout MindOS

周三晚 10:30，老板说："明天评审要看到能跑的项目骨架"

Wednesday 10:30 PM: "We need a runnable project skeleton for tomorrow's review."

1你反复说明：Next.js 15、pnpm、目录规范、CI 要求You repeatedly explain: Next.js 15, pnpm, folder conventions, CI requirements
2第一版用了 npm 且结构不符合团队约定First output uses npm and a structure that violates team conventions
3你二次纠偏后才能进入真正开发You spend another round correcting before real development can start

~25 min~25 min 启动慢，首版不可用slow kickoff, first output not usable

✓ 使用 MindOSWith MindOS

同一句话："帮我启动这个代码开发，明天评审要看"

Same ask: "Kick off this project build for tomorrow's review."

1Agent 读取 Profile/Identity.md → 默认技术栈与命令习惯Agent reads Profile/Identity.md for default stack and command preferences
2读取 Workflows/Startup-SOP.md → 自动带上初始化、校验、CI 模板Reads Workflows/Startup-SOP.md and auto-applies setup, checks, and CI template
3首版即可跑通，团队直接接力开发First output runs immediately; team can start building right away

~4 min~4 min 首版可用，直接进入迭代usable first output, straight into iteration

✕ 没有 MindOSWithout MindOS

你说："我今天和他聊了这些"

You say: "I talked with this person today."

1你重新解释人物背景、历史合作、敏感话题边界You re-explain relationship background, collaboration history, and sensitive boundaries
2本次会话记住了，但下次换 Agent 又要从头讲This session remembers, but switching agents means starting from zero again
3重要承诺容易遗漏，跟进动作断档Important commitments slip through, follow-up actions break

易断档Fragile 关系信息分散在会话里relationship context scattered across chats

✓ 使用 MindOSWith MindOS

你只说："我今天和他聊了这些，帮我推进下一步"

You only say: "We discussed this today, drive the next steps."

1Agent 从聊天里抽取关键事实与情绪变化，结构化记录Agent extracts key facts and sentiment shifts from the chat into structured notes
2不只更新 Connections/XXX.md，还自动生成跟进策略、待办和提醒时间Not just updates Connections/XXX.md, but also creates follow-up strategy, tasks, and reminder timing
3后续任何 Agent / 新会话都可直接接手执行，不再从头梳理关系脉络Any future agent/session can continue execution directly without rebuilding relationship history

可执行Actionable 从聊天到行动，自动闭环from conversation to execution, closed-loop

✕ 没有 MindOSWithout MindOS

周会后，产品、研发、运营都在各自工具里记录了"下一步"

After weekly sync, product, engineering, and ops all captured next steps in separate tools

1产品在文档写优先级，研发在代码工具记实现，运营在群里补执行计划Product tracks priorities in docs, engineering logs implementation notes, ops writes plans in chat
2信息分散且口径不一致，跨角色协作频繁二次确认Context is fragmented and inconsistent, forcing repeated cross-role confirmations
3一周后复盘才发现目标偏移，团队花时间补救对齐A week later, review reveals drift and the team spends time repairing alignment

团队失焦Team drift 每个人都在努力，但不在同一条线上everyone works hard, but not on one shared line

✓ 使用 MindOSWith MindOS

同样的周会结论，统一沉淀到团队 MindOS 里

The same meeting outcomes are captured into one team MindOS

1会议纪要自动更新 Team/Decisions.md、Projects/Roadmap.md、Workflows/Handoff-SOP.mdMeeting notes auto-update Team/Decisions.md, Projects/Roadmap.md, and Workflows/Handoff-SOP.md
2不同角色的 Agent 都读取同一份团队上下文，产出天然对齐Role-specific agents read the same team context, so outputs align by default
3每次任务推进都能追溯到团队决策，协作像同一个大脑在思考Each task traces back to team decisions, making collaboration feel like one shared brain

团队同频Team sync MindOS 成为团队共同思考层MindOS becomes the team's shared thinking layer

✕ 没有 MindOSWithout MindOS

发布前最后一小时，你说："Review 这个支付模块 PR"

One hour before release: "Review this payment-module PR."

1Agent 给出大量通用建议，但忽略你们的支付容错标准Agent gives many generic tips but misses your payment fault-tolerance standards
2你花时间筛噪音，真正高风险点被埋没You burn time filtering noise while real high-risk issues stay hidden
3上线后才暴露异常链路，回滚成本高Failure path shows up after release, forcing expensive rollback

高风险High risk 噪音多，关键点漏检high noise, critical misses

✓ 使用 MindOSWith MindOS

同一句话："Review 这个支付模块 PR"

Same ask: "Review this payment-module PR."

1Agent 读取 Configs/Code-Standards.md + 历史缺陷模式Agent reads Configs/Code-Standards.md plus historical defect patterns
2优先标出真正会导致事故的问题：幂等、回滚、超时兜底Prioritizes incident-prone issues: idempotency, rollback, timeout fallback
3输出按风险等级排序，开发可直接照单修复Outputs risk-ranked findings so engineers can fix immediately

高命中High signal 按团队标准命中关键风险critical risks found under your own standards

✕ 没有 MindOSWithout MindOS

同一个任务里，你在不同工具和新对话之间来回切换

Inside one task, you keep switching across tools and fresh chats

1每次一换工具或开新对话，就要重讲背景、约定和当前进度Every time you switch tools or start a new chat, you must re-explain context, conventions, and progress
2同一工具开新 session 也会丢记忆，回答风格和决策标准来回漂移Even a new session in the same tool loses memory, so style and decision criteria drift
3你在做"上下文搬运"，而不是推进结果交付You end up transporting context instead of shipping outcomes

高频重讲Frequent rebriefs 每次一换工具/对话，就要重新对齐every tool/chat switch forces re-alignment

✓ 使用 MindOSWith MindOS

同一个任务里，无论换工具还是开新对话，都从同一份 MindOS 上下文继续

In the same task, tool switches and new chats both continue from one shared MindOS context

1每个 Agent 通过 MCP 读取同一份 MindOS 知识库Every agent reads the same MindOS knowledge base via MCP
2偏好、标准、项目状态自动继承，新会话也保持一致判断Preferences, standards, and project state carry over automatically, even in fresh chats
3你只做决策与验收，协作链路持续不断流You stay on decisions and acceptance while collaboration flow remains continuous

0 重讲0 rebriefs 切换工具/对话，协作不断流switch tools/chats without breaking flow

05. FEATURES

核心功能特性

Core Features

人类侧For Humans

GUI 工作台

GUI Workbench

浏览、编辑、搜索笔记，统一搜索 + AI 入口（⌘K / ⌘/），专为人机共创设计。

Browse, edit, search notes with unified search + AI entry (⌘K / ⌘/), designed for human-AI co-creation.

内置 Agent 助手

Built-in Agent Assistant

在上下文中与知识库对话，Agent 管理文件，编辑无缝沉淀人类主动管理的知识。

Converse with the knowledge base in context. Agents manage files while editing seamlessly captures human-curated knowledge.

插件扩展

Plugin Extensions

多种内置渲染器插件 — TODO Board, CSV Views, Wiki Graph, Timeline, Agent Inspector 等，实现弹性知识管理。

Multiple built-in renderer plugins — TODO Board, CSV Views, Wiki Graph, Timeline, Agent Inspector, and more for elastic knowledge management.

Agent 侧For Agents

MCP Server & Skills

完整 MCP 工具集，stdio + HTTP 双传输，全阵容 Agent 兼容（OpenClaw, Claude Code, Cursor 等），零配置接入读写、搜索及执行本地工作流。

Full MCP tool surface, stdio + HTTP dual transport, full-lineup Agent compatible (OpenClaw, Claude Code, Cursor, etc.). Zero-config access to read, write, search, and execute local workflows.

结构化模板

Structured Templates

预置 Profile、Workflows、Configurations 等目录骨架，快速冷启动个人 Context。

Pre-set directory structures for Profiles, Workflows, Configurations, etc., to jumpstart personal context.

笔记即指令

Agent-Ready Docs

日常笔记天然就是 Agent 可直接执行的高质量指令——无需额外格式转换，写下即可调度。

Everyday notes naturally double as high-quality executable Agent commands — no format conversion needed, write and dispatch.

基础设施Infrastructure

安全防线

Security

Bearer Token 认证、路径沙箱、INSTRUCTION.md 写保护、原子写入——Agent 操作受限于安全边界内。

Bearer Token auth, path sandboxing, INSTRUCTION.md write-protection, atomic writes — Agent operations stay within secure boundaries.

可视化知识图谱

Visual Knowledge Graph

动态解析并可视化文件间的引用与依赖关系，直观管理人机上下文网络。

Dynamically parses and visualizes inter-file references and dependencies across the human-AI context network.

时光机 & 版本控制

Time Machine & Git-backed

Git 自动同步（commit/push/pull），记录人类与 Agent 的每次编辑历史，一键回滚，可视化 Context 演变。

Git auto-sync (commit/push/pull), records every edit by both humans and Agents. One-click rollback, visualize context evolution.

06. AGENT ECOSYSTEM

无缝链接 Agents 生态

Seamless Agent Ecosystem

OpenClaw 开源 Agent 框架Open-source Agent Framework

Claude Code 终端编程 AgentTerminal Coding Agent

Codex OpenAI 编程 AgentOpenAI Coding Agent

Gemini CLI Google 终端 AgentGoogle Terminal Agent

GitHub Copilot AI 编程副驾驶AI Pair Programmer

Cursor AI 原生编辑器AI-native Code Editor

Trae 字节 AI IDEByteDance AI IDE

CodeBuddy 腾讯 AI 编程助手Tencent AI Coding Assistant

OpenClaw 开源 Agent 框架Open-source Agent Framework

Claude Code 终端编程 AgentTerminal Coding Agent

Codex OpenAI 编程 AgentOpenAI Coding Agent

Gemini CLI Google 终端 AgentGoogle Terminal Agent

GitHub Copilot AI 编程副驾驶AI Pair Programmer

Cursor AI 原生编辑器AI-native Code Editor

Trae 字节 AI IDEByteDance AI IDE

CodeBuddy 腾讯 AI 编程助手Tencent AI Coding Assistant

真实体验REAL STORY

创造者的第一手体验

The Builder's First-Hand Experience

我同时使用 5 个 Agent——Claude Code 写代码、Cursor 做重构、Codex 补测试、Gemini CLI 做调研、Trae 写前端。每次切换工具，我都得重复说明：我的技术栈、代码风格、项目背景、命名约定……

I use 5 Agents daily — Claude Code for coding, Cursor for refactoring, Codex for tests, Gemini CLI for research, Trae for frontend. Every time I switch tools, I repeat the same context: my tech stack, code style, project background, naming conventions…

一个周末我统计了一下——一周里，我在不同 Agent 里重复了 47 次相同的上下文。每次都是复制粘贴同样的偏好，或者花 5 分钟解释"我们这个项目用 pnpm 不用 npm"。

One weekend I counted — in a single week, I repeated the same context 47 times across different Agents. Copy-pasting the same preferences, or spending 5 minutes explaining "this project uses pnpm, not npm."

于是我把所有偏好和工作流写进一个本地知识库，通过 MCP 协议让每个 Agent 自动读取。第一个 Agent 搭项目时自动用了 Next.js 15 + Tailwind + pnpm；第二个做审查时按我的规范提 PR；第三个写文档时用了我的语气风格。

So I wrote all my preferences and workflows into a local knowledge base, exposed via MCP for every Agent to read. The first Agent scaffolded with Next.js 15 + Tailwind + pnpm automatically; the second reviewed PRs against my standards; the third wrote docs in my voice.

47 次重复 → 0 次。不是效率优化——是体验的质变。我终于不再当 Agent 的"人肉上下文复读机"了。

47 repetitions → 0. Not an optimization — a paradigm shift. I finally stopped being a "human context copy-paste machine" for my Agents.

47× 每周重复上下文 weekly context repeats

→

0 使用 MindOS 后 with MindOS

→

5+ Agent 共享同一心智 Agents share one mind

GeminiLight · MindOS 创造者

GeminiLight · Creator of MindOS

AI 独立开发者 · 同时使用 5+ Agent 的重度用户

AI Indie Developer · Heavy user of 5+ Agents daily

07. QUICKSTART

立即安装

Install Now

选择你的方式

Choose Your Way

macOS ARM64 macOS Intel Windows x64 Linux x64

macOS 安全提示macOS Security Note

首次打开如遇提示，请运行以下命令： If blocked on first launch, run this command:

xattr -cr /Applications/MindOS.app 点击复制命令Click to copy

CLI

                            npm i -g @geminilight/mindos && mindos onboard
                        

安装Install → 初始化模板Init template → 配置环境Configure env → 注册 MCPRegister MCP → 安装 SkillsInstall Skills

—— 或发给你的 Agent ———— or send to your Agent ——

安装 PromptInstall Prompt

帮我从 https://github.com/GeminiLight/MindOS 安装 MindOS，包含 MCP 和 Skills，使用中文模板。 Help me install MindOS from https://github.com/GeminiLight/MindOS with MCP and Skills. Use English template.

Claude Code Cursor Cline Windsurf CodeBuddy Trae Gemini CLI

2

安装完成？试试这些

Installed? Try These

直接粘贴给 Agent，立即体验 MindOS 的核心能力。

Paste any of these to your Agent and experience MindOS instantly.

👤

注入身份

Inject Profile

这是我的简历，读一下，把我的信息整理到 MindOS 里。

Here's my resume, read it and organize my info into MindOS.

🔄

沉淀经验

Distill SOP

帮我把这次对话的经验沉淀到 MindOS，形成一个可复用的工作流。

Help me distill the experience from this conversation into MindOS as a reusable SOP.

▶️

执行工作流

Run Workflow

帮我执行 MindOS 里的 XXX 工作流。

Help me execute the XXX SOP from MindOS.

人机共享心智

Human-AI Shared Mind

全局同步 — 打破记忆割裂

Global Mind Sync — Breaking Memory Silos

透明可控 — 消除记忆黑箱

Transparent & Controllable — No Black Boxes

共生演进 — 经验回流为指令

Symbiotic Evolution — Experience Flows Back as Instructions

底层基石：本地优先

Foundational Pillar: Local-first

想法、执行到复盘，一条线贯穿

From Ideas to Execution to Review, One Thread Through All

核心思路

Key Approach

关联更新

Related Updates

产品一览See It in Action

首页Home

AI 对话AI Chat

Agent 工作台Agents Dashboard

EchoEcho

同一个任务，两种体验

Same Task, Two Realities

交互式心智循环

Interactive Mind Loop

人类心智

Human Mind

Agent 舰队

Agent Fleet

核心功能特性

Core Features

GUI 工作台

GUI Workbench

内置 Agent 助手

Built-in Agent Assistant

插件扩展

Plugin Extensions

MCP Server & Skills

MCP Server & Skills

结构化模板

Structured Templates

笔记即指令

Agent-Ready Docs

安全防线

Security

可视化知识图谱

Visual Knowledge Graph

时光机 & 版本控制

Time Machine & Git-backed

无缝链接 Agents 生态

Seamless Agent Ecosystem

创造者的第一手体验

The Builder's First-Hand Experience

立即安装

Install Now

安装完成？试试这些

Installed? Try These