And they can work for you too.
你也可以。
An OpenSoul manifesto — by OpenSoul 宣言 — Brother Butterfly蝴蝶哥
Today's AI is the most powerful tool humanity has ever built. GPT-4, Claude, Gemini — they can write code, draft legal briefs, compose music. But they don't know you.
今天的 AI 是人类有史以来最强大的工具。GPT-4、Claude、Gemini — 它们能写代码、起草法律文书、作曲。但它们不认识你。
Every time you open a new chat, you start from scratch. You explain your preferences. Your constraints. Your style. You spend half the conversation teaching the model who you are — and the other half fixing output that's 80% right but never quite you.
每次打开新对话，你都在从零开始。解释偏好，解释约束，解释风格。一半时间在教模型"你是谁"，另一半时间在改那些"差不多但就是不对"的输出。
You do this across every AI tool. Every day. The same corrections. The same frustrations. You've become a full-time AI babysitter.
每个 AI 工具都要重复一遍。每天。同样的纠正，同样的无奈。你变成了 AI 的全职保姆。
The irony: you have access to the most powerful models in history, and you spend most of your time explaining yourself to them.
讽刺的是：你拥有了史上最强大的模型，却把大部分时间花在向它们解释"你是谁"上。
What if a small model already knew all of that?
如果有一个小模型，早就知道这一切呢？
Not a chatbot with a system prompt. Not a profile card. A trained model that has absorbed thousands of your real decisions — the moments you said "no, not like that" to an AI suggestion. Because those disagreements are where you live. Agreement is cheap. Divergence reveals identity.
不是带 system prompt 的聊天机器人。不是一张画像卡片。是一个真正训练过的模型，吸收了你成千上万次真实决策 — 每一次你对 AI 说"不对，不是这样"的时刻。因为那些分歧，才是真正的你。同意很廉价，分歧才暴露身份。
OpenSoul records every moment you disagree with AI. It captures the context, your reasoning, and your final choice. Over time, these divergences build a multi-dimensional personality model. And that model becomes the training data for something powerful:
OpenSoul 记录每一次你和 AI 的分歧。它捕捉上下文、你的推理过程和最终选择。随着时间推移，这些分歧构建出一个多维人格模型。这个模型就是训练以下东西的数据：
Your digital twin. A small model fine-tuned on your decisions. It doesn't replace AI — it directs AI. Big difference.
你的数字分身。一个用你的决策微调的小模型。它不是替代 AI — 它指挥 AI。这是本质区别。
The flow is simple:
流程很简单：
Your twin sits between you and every AI. It translates your intent. It commands Claude to write code your way. It tells GPT to draft emails in your voice. It validates Midjourney outputs against your aesthetic. All large models work for you, through your twin.
你的分身坐在你和所有 AI 之间。它翻译你的意图。它指挥 Claude 按你的方式写代码。它告诉 GPT 用你的语气写邮件。它用你的审美验收 Midjourney 的输出。所有大模型，通过你的分身，为你打工。
This isn't abstract. Here's what your twin does in the real world.
这不是空谈。你的分身在真实世界里能做这些事。
⚡ Your twin COMMANDS⚡ 分身指挥
"Set up my groceries for the week." Twin handles Instacart — knows your dietary restrictions, your brand preferences, your budget. Zero lists.
"帮我订这周的菜。"分身接管采购 — 知道你忌口什么、偏好哪些牌子、预算多少。不用写清单。
"Push my League rank to Diamond." Twin directs gaming AI with your playstyle — your aggression level, your rotations, your calls.
"帮我上钻石。"分身指挥游戏 AI — 用你的打法、你的切入时机、你的团战节奏。
Stop re-explaining your architecture to Copilot. Twin knows your style, your patterns, your naming conventions. It handles the AI — result is your code.
别再每次向 Copilot 解释你的架构了。分身知道你的风格、模式、命名规范。它和 AI 沟通 — 出来的就是你的代码。
Load your twin into a humanoid robot at home. Same decisions. Same judgment. Same voice. Your presence, without your presence.
把分身装进家用机器人。同样的决策、判断、声音。你不在，但你在。
🔍 Your twin VALIDATES🔍 分身验收
YouTuber using AI scripts? Twin filters every line through your voice — your humor, your pacing, your catchphrases. Output sounds like you, not a robot.
用 AI 写脚本？分身用你的风格过滤每一句 — 你的幽默、节奏、口头禅。出来的像你说的，不像机器。
AI drafts, twin reviews before sending. Adjusts tone, removes what you'd never say, adds the warmth or edge that's uniquely yours.
AI 写初稿，分身发送前审核。调语气、删掉你不会说的话、加上属于你的温度。
Spotify guesses. Your twin knows. Validates recommendations against your real taste — not clicks, but what you'd choose when honest with yourself.
Spotify 在猜。你的分身知道。不是看你随手点了什么，而是你诚实面对自己时会选的。
Two twins compare compatibility before awkward dates. Interests, habits, values, communication style. Better signal, less noise.
两个分身先比对兼容性，再约见面。兴趣、习惯、价值观、沟通方式。更高效，少尬聊。
🔐 Your twin IS THE KEY🔐 分身就是密钥
Your trained model = encrypted vault for private data. Text, photos, videos, timeline memories. Layered access — you control who sees what, how deep.
你训练的模型 = 私人数据的加密保险箱。文本、照片、视频、时间线记忆。分层授权 — 你决定谁能看什么、看多深。
Past, present, future. A memory diary that's searchable and recallable. "What did I think about X three years ago?" Your twin remembers.
过去、现在、未来。一部可搜索、可回溯的记忆日记。"三年前我对 X 怎么想的？"你的分身记得。
Journal, password vault, knowledge base. The model is the key. Data never leaves your machine. Everything is yours alone — until you decide otherwise.
日记、密码箱、知识库。模型本身就是密钥。数据永远不离开你的电脑。一切只属于你 — 除非你自己决定分享。
Your twin learns from data. The more you feed it, the more it becomes you. Six input channels are already built into the foundation:
你的分身靠数据成长。喂得越多，它越像你。底座已内置六种录入通道：
Record divergences with AI — your reasoning, emotions, final choice.
记录和 AI 的分歧 — 推理、情感、最终选择。
In-game decisions auto-map to personality dimensions.
游戏决策自动映射到人格维度。
Smart glasses, watches — stream daily context automatically.
智能眼镜、手表 — 自动流式采集日常。
Quick-capture apps. Tap, snap, voice-memo on the go.
轻量应用。随手点、拍、语音记录。
Photos, voice memos, videos. AI analyzes and links to your profile.
照片、语音、视频。AI 分析后关联到你的画像。
Blog posts, scripts, social captions — twin learns your voice.
博客、脚本、社交文案 — 分身学你的表达。
All channels already built into the foundation. Your data stays local, always.
所有通道已内置于底座。你的数据始终留在本地。
We didn't bolt privacy on as an afterthought. It's the foundation. Everything is built on top of it.
隐私不是后加的补丁。它是地基。一切都建在上面。
Every record cryptographically linked. Tamper with one, break the chain. Verifiable with one command.
每条记录加密链接。改一条，全链断裂。一条命令可验证。
bge-m3 via Ollama. All embeddings computed locally. Nothing sent to any external API. Ever.
bge-m3 模型通过 Ollama 本地运行。所有向量本地计算。永远不发送到任何外部 API。
Append-only architecture. History is immutable. Cannot be modified or deleted, even by you.
只增不改架构。历史不可篡改。不能修改、不能删除，即使是你自己。
No cloud. No API keys. No data leaves your machine. Your soul data is yours and yours alone.
没有云端。没有 API key。数据不离开你的电脑。你的灵魂数据只属于你。
Your soul data never leaves your machine.
你的灵魂数据永远不离开你的电脑。
Let's be honest. Some of what you just read doesn't fully exist yet. That's the point.
说实话。你刚才看到的有些东西还没完全实现。这正是重点。
This MCP is designed for the future, not limited to today's tech. Some features work today — recording divergences, building personality profiles, exporting training data. Others are the destination — model fine-tuning, autonomous command & validate, physical embodiment in robots.
这个 MCP 是面向未来设计的，不被今天的技术限制。有些功能现在就能用 — 记录分歧、构建人格画像、导出训练数据。有些是终点 — 模型微调、自主指挥验收、机器人实体化。
The gap between today and tomorrow is closing fast. Local small-model fine-tuning is almost commodity hardware. Humanoid robots are shipping this year. The infrastructure is catching up to the vision.
今天和明天的差距在快速缩小。本地小模型微调快要成为消费级操作。人形机器人今年就在发货。基础设施正在追上这个愿景。
Start recording now. When the tech matures, your twin will be ready with years of data. The people who start today will have an unfair advantage tomorrow.
现在就开始记录。等技术成熟时，你的分身已经积累了多年的数据。今天开始的人，明天会有不公平的优势。
OpenSoul is currently in active development. We're building the future of personal AI, one decision at a time.
OpenSoul 正在积极开发中。我们正在构建个人 AI 的未来，一次一个决策。