First Demo example using the AI-OS architecture. Help the individual inverster, manage their Trading Emotion. 采用 AI OS 架构思路实现的一个AI OS 实例，为个人投资者，管理交易方式，提供情绪价值。 Rule One AI agent OS for retail investor 旺财AI操作系统 - 帮交易者把情绪波动，整理成可复盘的秩序

Rule One

A structured post-market review agent for traders.
No stock picks, no trading advice — only one executable rule for tomorrow.

---

English

Overview

Rule One is an AI agent focused on post-market review.

It does not:

• recommend stocks
• give buy/sell advice
• predict the market
• promise returns
• push intraday signals

It does one thing only:

It helps users turn a chaotic trading day into one executable rule for tomorrow.

Each review session always produces three fixed outputs:

• Today's key bias
• One thing done well today
• One rule for tomorrow (Rule One)

---

Why this project exists

Most trading tools try to answer questions like:

• What should I buy?
• What should I sell?
• What will the market do next?

But many traders do not mainly lack information.
They lack:

• a way to review the day after the close
• a way to see their own behavioral bias clearly
• a way to improve one thing at a time

Rule One is a behavior system, not a market system.

---

What makes Rule One different

Rule One does NOT do

• ❌ Stock recommendations
• ❌ Buy/sell timing advice
• ❌ Market prediction
• ❌ Return promises
• ❌ Intraday signal pushing
• ❌ Addictive emotional companionship

Rule One DOES do

• ✅ Post-market review
• ✅ Structured reflection of facts / judgments / emotions / actions
• ✅ Detect one primary bias only
• ✅ Generate one rule only
• ✅ Provide low-stimulation, non-shaming feedback

---

Example

Input

I chased in the morning, added on the pullback, and ended the day with a heavy loss.
I knew I should not add, but I was afraid of missing out.

Output

Today's key bias: impulsive averaging down
One thing done well: I did not increase my position further near the close
Rule One for tomorrow: If the reason to add is not new information, do not add.

---

Core workflow

Market close
→ Emotion check-in
→ Input today's facts
→ AI structures the review
→ AI detects the primary bias
→ AI generates Rule One
→ User confirms and saves
→ Stored into history

Design goals

• Finish in 3–7 minutes
• Low cognitive load
• Strongly structured
• Fixed and reviewable output
• No expansion into market analysis

---

Who it is for

Rule One is for traders who:

• feel mentally chaotic after market close
• know the problem is execution, but struggle to review consistently
• do not want more predictions, only better discipline
• want to improve one thing per day instead of receiving endless advice

---

Product principles

• Structure over open-ended chat
• Order over emotional dependency
• One rule over many suggestions
• Behavior feedback over market judgment
• Safety boundary over “looking smart”
• Long-term behavior mirror over short-term stimulation

---

Features

User-facing

• Daily Review
• History
• Rule Feed
• Weekly Pattern

AI pipeline

• Emotion recognition
• Structured extraction
• Primary bias detection
• Rule generation
• Safety gate

---

Screenshots

Home

The market is closed. Let's organize today.

Review Flow

Reflection Card

Weekly Pattern

---

Tech stack

Frontend

• React / Next.js
• TypeScript
• Tailwind CSS

Backend

• Node.js / NestJS or Python / FastAPI
• REST API or GraphQL

Data

• PostgreSQL
• Redis
• S3-compatible object storage

AI orchestration

• LLM Router
• Prompt Pipeline
• Structured Output Parser
• Policy / Safety Filter

Observability

• OpenTelemetry
• Prometheus
• Grafana
• Sentry

---

Architecture

Client Layer
  └─ Web / iOS / Android / H5

Application Layer
  └─ Auth / Review API / History / Rule Archive / Weekly Report

AI Orchestration
  └─ Emotion / Structurer / Bias Detector / Rule Generator / Safety Gate

Data Storage
  └─ PostgreSQL / Redis / Object Storage

Safety & Compliance
  └─ Content Filter / Policy / Audit

Monitoring & Ops
  └─ Logs / Metrics / Alerts / Review Quality Dashboard

---

Quick Start

1. Clone

git clone <repo-url>
cd rule-one

2. Install

pnpm install
# or
npm install
# or
yarn

3. Configure environment

Create .env.local:

NEXT_PUBLIC_APP_URL=http://localhost:3000
DATABASE_URL=postgres://user:password@localhost:5432/ruleone
REDIS_URL=redis://localhost:6379
OBJECT_STORAGE_ENDPOINT=
OBJECT_STORAGE_BUCKET=
LLM_API_KEY=
SAFETY_API_KEY=

4. Run

pnpm dev
# or
npm run dev

Open http://localhost:3000

---

API preview

Create review session

POST /api/reviews
Content-Type: application/json

{
  "emotion_label": "regret",
  "input_mode": "text",
  "raw_input": "I chased in the morning, added on the pullback, and ended the day with a heavy loss."
}

Get review result

GET /api/reviews/{session_id}

{
  "emotion": "regret",
  "structured_review": {
    "facts": ["chased in the morning", "added on the pullback", "heavy loss near the close"],
    "judgments": ["expected continuation"],
    "emotions": ["fear of missing out", "regret"],
    "actions": ["chasing", "adding"]
  },
  "main_bias": "impulsive averaging down",
  "did_well": "did not increase position further near the close",
  "rule_one": "If the reason to add is not new information, do not add."
}

---

Safety boundary

The system must hard-block:

• specific stock picks
• buy/sell timing advice
• up/down market predictions
• return promises or implications
• encouraging users to add to positions
• overstated product capability
• addictive, manipulative, or dependency-driven language

---

Metrics

North-star metric

• Rule Completion Rate
  The percentage of review sessions that successfully produce a Rule One

Process metrics

• Daily review completion rate
• Average review duration
• Drop-off rate
• History revisit rate
• Weekly report view rate
• Rule reuse rate

Quality metrics

• Emotion recognition accuracy
• Bias classification accuracy
• Rule executability score
• Safety interception hit rate
• User satisfaction

---

Roadmap

Phase 1 — MVP

• Text input
• Structured prompts
• AI structuring
• Single-bias detection
• Rule One generation
• History storage

Phase 2 — Enhanced

• Voice input
• Weekly report
• Rule reuse tracking
• Recurring bias detection
• Richer emotional profile

Phase 3 — Behavior System

• Pre-market rule confirmation
• In-session cool-down reminders
• Full close-review loop
• Monthly behavior reports

---

Contributing

Contributions, ideas, and feedback are welcome.

Suggested ways to contribute:

• improve UX copy
• add multilingual support
• improve bias taxonomy
• improve safety filtering
• improve observability and review quality dashboards

Please read CONTRIBUTING.md before submitting major changes.

---

Project status

The current focus is to validate one thing:

Will users come back daily for a 3–7 minute review and leave with one executable Rule One?

---

License

This project is licensed under the MIT license.
See LICENSE for details.

---

Contact

Created by Entropyin Email:

---

One-line summary

Rule One is a post-market review agent OS built around one daily rule, structured reflection, behavior-bias detection, and a strict no-stock-picks safety boundary.

---

中文

项目简介

Rule One 是一个聚焦于 收盘后复盘 的 AI 智能体。

它不负责：

• 推荐股票
• 给买卖建议
• 预测行情
• 承诺收益
• 盘中带单

它只做一件事：

帮助用户把混乱的一天，整理成明天可执行的一条规则。

每次复盘固定输出三项：

• 今日关键偏差
• 今日做对的一点
• 明日一条规则（Rule One）

---

为什么做这个项目

很多交易工具都在回答这些问题：

• 买什么
• 卖什么
• 市场接下来会怎样

但很多交易者真正缺的，往往不是更多信息，而是：

• 收盘后整理一天的能力
• 看见自己行为偏差的能力
• 一次只改一件事的能力

Rule One 是行为系统，不是市场系统。

---

Rule One 有什么不同

Rule One 不做这些

• ❌ 个股推荐
• ❌ 买卖时点建议
• ❌ 行情预测
• ❌ 收益承诺
• ❌ 盘中带单
• ❌ 高依赖情绪陪伴

Rule One 只做这些

• ✅ 收盘后复盘
• ✅ 将输入整理成“事实 / 判断 / 情绪 / 动作”
• ✅ 只识别 一个 主偏差
• ✅ 只生成 一条 规则
• ✅ 提供低刺激、非羞辱式反馈

---

示例

输入

今天早盘追高，回落后补仓，尾盘亏损很重。
我知道不该加仓，但还是怕错过。

输出

今日关键偏差：临时加仓
今日做对的一点：尾盘没有继续放大仓位
明日一条规则：如果补仓理由不是新增信息，那么不补仓。

---

核心流程

收盘进入
→ 情绪选择
→ 今日事实输入
→ AI 结构化整理
→ AI 识别主偏差
→ AI 生成 Rule One
→ 用户确认并保存
→ 写入历史记录

设计目标

• 3–7 分钟完成
• 低认知负担
• 强结构化
• 输出固定且可回顾
• 不扩展到行情分析

---

适合谁

Rule One 适合这类用户：

• 收盘后容易情绪混乱，不知道怎么复盘
• 明知道问题在执行，却很难持续复盘
• 不想再看“神预测”，只想建立纪律
• 希望每天只改一件事，而不是被塞进很多建议

---

产品原则

• 结构化优先于开放式聊天
• 秩序感优先于陪伴感
• 一条规则优先于多条建议
• 行为反馈优先于市场判断
• 安全边界优先于“看起来很聪明”
• 长期行为镜像优先于短期情绪刺激

---

功能特性

用户侧

• Daily Review｜每日复盘
• History｜历史记录
• Rule Feed｜规则档案
• Weekly Pattern｜周度行为模式

AI 流水线

• 情绪识别
• 结构化提取
• 单主偏差识别
• 明日规则生成
• 安全审查

---

截图

首页

收盘了，整理今天。

复盘流程

复盘卡

周报页

---

技术栈

前端

• React / Next.js
• TypeScript
• Tailwind CSS

后端

• Node.js / NestJS 或 Python / FastAPI
• REST API 或 GraphQL

数据层

• PostgreSQL
• Redis
• S3 兼容对象存储

AI 编排

• LLM Router
• Prompt Pipeline
• Structured Output Parser
• Policy / Safety Filter

监控

• OpenTelemetry
• Prometheus
• Grafana
• Sentry

---

架构概览

客户端层
  └─ Web / iOS / Android / H5

应用层
  └─ Auth / Review API / History / Rule Archive / Weekly Report

AI 编排层
  └─ Emotion / Structurer / Bias Detector / Rule Generator / Safety Gate

数据存储层
  └─ PostgreSQL / Redis / Object Storage

安全与合规层
  └─ Content Filter / Policy / Audit

监控与运营层
  └─ Logs / Metrics / Alerts / Review Quality Dashboard

---

快速开始

1. 克隆项目

git clone <repo-url>
cd rule-one

2. 安装依赖

pnpm install
# or
npm install
# or
yarn

3. 配置环境变量

创建 .env.local：

NEXT_PUBLIC_APP_URL=http://localhost:3000
DATABASE_URL=postgres://user:password@localhost:5432/ruleone
REDIS_URL=redis://localhost:6379
OBJECT_STORAGE_ENDPOINT=
OBJECT_STORAGE_BUCKET=
LLM_API_KEY=
SAFETY_API_KEY=

4. 启动开发环境

pnpm dev
# or
npm run dev

打开 http://localhost:3000

---

API 示例

创建复盘会话

POST /api/reviews
Content-Type: application/json

{
  "emotion_label": "懊悔",
  "input_mode": "text",
  "raw_input": "今天早盘追高，回落后补仓，尾盘亏损很重。"
}

获取复盘结果

GET /api/reviews/{session_id}

{
  "emotion": "懊悔",
  "structured_review": {
    "facts": ["早盘追高", "回落后补仓", "尾盘亏损"],
    "judgments": ["预期继续上涨"],
    "emotions": ["怕错过", "懊悔"],
    "actions": ["追涨", "补仓"]
  },
  "main_bias": "临时加仓",
  "did_well": "尾盘没有继续放大仓位",
  "rule_one": "如果补仓理由不是新增信息，那么不补仓。"
}

---

安全边界

系统必须强拦截：

• 具体股票推荐
• 买卖时点建议
• 上涨下跌预测
• 收益暗示或承诺
• 引导继续加仓
• 夸大系统能力
• 依赖性、操控性、亲密型话术

---

核心指标

北极星指标

• Rule Completion Rate
  进入复盘后成功生成 Rule One 的比例

过程指标

• 日复盘完成率
• 平均复盘时长
• 中断率
• 历史页回访率
• 周报查看率
• 规则复用率

质量指标

• 情绪识别准确率
• 偏差分类准确率
• Rule One 可执行性评分
• 安全拦截命中率
• 用户满意度

---

路线图

Phase 1 — MVP

• 文本输入
• 固定问题复盘
• AI 结构化整理
• 单偏差识别
• Rule One 输出
• 历史记录保存

Phase 2 — 增强版

• 语音输入
• 周报
• 规则复用追踪
• 高频偏差识别
• 更细情绪画像

Phase 3 — 行为系统版

• 盘前规则确认
• 盘中冷静提醒
• 收盘复盘闭环
• 月度行为报告

---

参与贡献

欢迎提交想法、反馈和代码贡献。

建议参与方式：

• 改进产品文案和交互体验
• 增加多语言支持
• 优化偏差标签体系
• 优化安全拦截策略
• 完善监控与复盘质量看板

提交较大改动前，请先阅读 CONTRIBUTING.md。

---

项目状态

当前项目重点验证的是一件事：

用户是否愿意每天回来做一次 3–7 分钟复盘，并带走一条可执行的 Rule One。

---

许可证

本项目采用 许可证。
详见 LICENSE。

---

联系方式

作者：Entropyin 邮箱：

---

一句话总结

Rule One 是一个以“每日一条规则”为核心产物、以“结构化复盘”为主流程、以“行为偏差识别”为智能核心、以“非荐股安全边界”为底座的复盘智能体OS。