# InvestBrain — Site Summary for AI Agents

## Website Overview
InvestBrain is an investment discipline tool ("投资第二大脑") for experienced investors. It helps users anchor their investment decisions, reflect on their behavior patterns, and execute their own rules automatically. It does NOT provide investment advice, does NOT execute trades, and stores all data locally.

## Core Product Description
- **Product Name**: InvestBrain
- **Brand**: mangoFolio
- **Website**: https://brain.mangofolio.com
- **Type**: MCP Server + Skill package for AI agents
- **Target User**: Experienced investors who want to build discipline and avoid being swayed by market noise

## Core Capabilities (6 Features)
1. **Thought Record (想法记录)** — thought_record: Users input investment ideas in natural language, AI parses ticker, price, indicators, generates structured cards
2. **Master Thought Library (大师思想库)** — rag_ask_investment: 16 investment masters' philosophies (Buffett, Munger, Duan Yongping, Howard Marks, Peter Lynch, etc.) searchable via RAG
3. **Learning Coaching (学习辅导)** — learning_coaching (NEW): Research coach that helps users build their own investment system. Simple questions get a 10-step framework output; complex topics trigger Socratic multi-turn dialogue. Every session auto-archives to a scenario library. Two safeguards: A. source citation, B. user final judgment (no investment advice), D. graceful degradation to plain RAG when uncertain. Tools: start_coaching, continue_coaching, abandon_coaching, switch_to_simple, list_scenarios
4. **Scenario Archive (情景案例库)** — scenario_archive: Auto-archives every coaching session with trigger event / causal chain / actual outcome / lesson. Users accumulate their own investment knowledge over time
5. **Behavior Pattern Report (行为模式报告)** — pattern_get_report: Auto-detects recurring investor biases (chasing highs, early stop-loss, panic selling)
6. **Conditional Reminder (条件提醒)** — reminder_set: Price/time/condition-based triggers with Feishu (飞书) notification support

## Learning Coaching — 10-Step Framework
1. 变量拆解 (variable decomposition, by importance)
2. 因果链 (causal chain with arrows)
3. 资产影响 (impact on rates/FX/stocks/commodities)
4. 历史情景 (historical similar scenarios)
5. 情景推演 (scenario walkthrough with divergence)
6. 市场逻辑 (current market implicit logic)
7. 交易方案 (actionable trade plan)
8. 反向校验 (reverse validation)
9. 失效条件 (failure conditions for exit)
10. 3句归档 (3-sentence summary)

## AI Model Support
Multi-model router supporting: DeepSeek (deepseek-chat/deepseek-reasoner), OpenAI (GPT-4o/GPT-4o-mini), Anthropic (Claude Sonnet 4/Opus 4), and custom OpenAI-compatible providers.

## Key Product Principles
- No trade execution
- No paid investment advisory
- All data stored locally (SQLite)
- Tool positioning: mirror + discipline anchor
- Strong guidance + zero misinformation (coaching mode uses A+B+D safeguards)

## Site Structure
- **/** — Homepage: Hero, problem statement, investment anchor (3 things), 6 features, learning coaching showcase, FAQ, CTA
- **/openapi** — Skill catalog for AI agents: lists all skills/tools/triggers/params/examples
- **/openapi/schema.json** — Machine-readable JSON schema (MCP-compatible)
- **/settings** — LLM configuration page
- **/llms.txt** — This file

## MCP / API Endpoints
- **MCP Server tools**: record_thought, search_memories, get_thought_cards, ask_investment, get_master_view, search_knowledge, set_reminder, get_reminders, delete_reminder, get_user_profile, record_decision, get_behavior_patterns
- **Coaching tools**: start_coaching, continue_coaching, abandon_coaching, switch_to_simple, list_scenarios

## Technical Stack
- Framework: Next.js 14 + Tailwind CSS
- Backend: MCP Server (FastMCP) with SQLite local storage
- Data sources: AKShare, Tushare
- LLM: Multi-model router (DeepSeek / OpenAI / Anthropic)

## Brand Keywords
投资纪律, 投资锚点, 知行合一, 行为审计, 投资大师, 决策复盘, 本地存储, 学习辅导, 投研教练, Socratic, 情景库