🤖 StockAgent — LLM-Powered Multi-Agent Stock Analysis System

An intelligent stock analysis system that combines multi-source data fusion, multi-dimensional scoring, and LLM-powered reasoning to provide actionable market insights. Designed as a multi-agent architecture where specialized agents handle data collection, analysis, screening, and reporting.

🏗️ Architecture

┌─────────────────────────────────────────────────────────────────┐
│                     User Interface Layer                        │
│            CLI • WeChat Bot • Web Dashboard • API               │
└──────────────────────────┬──────────────────────────────────────┘
                           │
┌──────────────────────────▼──────────────────────────────────────┐
│                     Agent Orchestrator                          │
│         Task Scheduling • Memory • Context Management           │
└───┬──────────┬──────────┬──────────┬──────────┬─────────────────┘
    │          │          │          │          │
┌───▼───┐ ┌───▼───┐ ┌───▼───┐ ┌───▼───┐ ┌───▼───┐
│ Data  │ │Analysis│ │Screen │ │Report │ │Monitor│
│ Agent │ │ Agent  │ │ Agent │ │ Agent │ │ Agent │
│       │ │        │ │       │ │       │ │       │
│Multi- │ │Technic.│ │Full   │ │Daily  │ │Real-  │
│Source │ │Capital │ │Market │ │Brief  │ │time   │
│Fusion │ │Sentim. │ │Scan   │ │Gener. │ │Alert  │
│       │ │Fundam. │ │17+    │ │LLM    │ │       │
│5+ API │ │Market  │ │Factor │ │Powered│ │Cron   │
└───┬───┘ └───┬───┘ └───┬───┘ └───┬───┘ └───┬───┘
    │         │         │         │         │
┌───▼─────────▼─────────▼─────────▼─────────▼────────────────────┐
│                    Data Source Layer                            │
│  Tencent • Sina • EastMoney • THS • AkShare • Tushare • YFinance│
│  Fallback Chain • Cache • Rate Limiting                        │
└────────────────────────────────────────────────────────────────┘

✨ Key Features

🔄 Multi-Source Data Fusion with Fallback Chain

Real-time quotes from 5+ providers (Tencent, Sina, EastMoney, THS, AkShare) with automatic failover. If one source is down, the system seamlessly switches to the next.

# Data sources with automatic fallback
sources = [TencentRealtimeSource(), SinaRealtimeSource(), EastMoneyRealtimeSource()]
chain = FallbackChain(sources)
quote = await chain.get_quote("002475")  # Tries each source until one succeeds

📊 Multi-Dimensional Scoring Engine (17+ Factors)

Every stock gets scored across 5 dimensions:

Dimension	Factors	Weight
Technical	MA trends, RSI, consecutive candles, price position	25%
Capital Flow	Main force net flow, outer/inner ratio, volume ratio	25%
Sentiment	News polarity, market heat, social media signals	20%
Fundamental	PE, PB, revenue growth, ROE	20%
Market	Sector strength, index correlation	10%

🔍 Full Market Screening

Scan 5000+ A-share stocks in seconds with configurable filters:

• Bowlong anti-chase rules (avoid overbought stocks)
• Fundamental risk gating (ST, delisting, insider selling)
• Event-driven scoring (news catalysts)
• Holdings overlap detection

📰 LLM-Powered Analysis

Feed market data to LLM for natural language analysis:

• Daily morning/evening reports
• Stock-specific deep dives
• Portfolio risk assessment
• Market sentiment summary

⏰ Automated Monitoring

• Scheduled market scans (pre-market, mid-day, post-market)
• Real-time price alerts via WeChat/Telegram
• Portfolio tracking with stop-loss monitoring

🚀 Quick Start

# Install
pip install -e .

# Single stock analysis
stock-agent analyze 002475

# Full market scan
stock-agent scan --top 20

# Market overview
stock-agent market

# Generate daily report
stock-agent report morning

📁 Project Structure

stock-agent/
├── src/stock_agent/
│   ├── __init__.py
│   ├── __main__.py          # CLI entry point
│   ├── tools/               # Agent tools (callable by LLM)
│   │   ├── stock_analysis.py
│   │   ├── market_scan.py
│   │   ├── daily_report.py
│   │   └── price_alert.py
│   ├── data/                # Multi-source data layer
│   │   ├── sources/         # Data source adapters
│   │   │   ├── base.py      # Abstract base + FallbackChain
│   │   │   ├── tencent.py
│   │   │   ├── sina.py
│   │   │   ├── eastmoney.py
│   │   │   └── ths.py
│   │   ├── manager.py       # Unified data manager
│   │   └── cache.py         # TTL cache layer
│   ├── analysis/            # Scoring & analysis engine
│   │   ├── scoring.py       # Multi-dimensional scoring
│   │   ├── technical.py     # Technical indicators
│   │   ├── capital_flow.py  # Capital flow analysis
│   │   ├── sentiment.py     # News sentiment (LLM-powered)
│   │   └── fundamental.py   # PE/PB/ROE analysis
│   ├── agent/               # LLM agent integration
│   │   ├── tools.py         # OpenAI function-calling schemas
│   │   ├── prompts.py       # System prompts
│   │   └── orchestrator.py  # Multi-agent coordinator
│   └── reporting/           # Report generation
│       ├── morning.py       # Pre-market brief
│       ├── evening.py       # Post-market summary
│       └── templates.py     # Report templates
├── tests/
│   ├── test_scoring.py
│   ├── test_data_sources.py
│   └── test_technical.py
├── pyproject.toml
├── README.md
└── LICENSE

🔧 Agent Tools (LLM Function Calling)

Each tool is designed to be called by an LLM agent via function calling:

# tools/stock_analysis.py
TOOLS = [
    {
        "type": "function",
        "function": {
            "name": "stock_analysis",
            "description": "Analyze a single stock with multi-dimensional scoring",
            "parameters": {
                "type": "object",
                "properties": {
                    "code": {"type": "string", "description": "Stock code, e.g. '002475'"},
                    "include_sentiment": {"type": "boolean", "default": True},
                },
                "required": ["code"]
            }
        }
    },
    {
        "type": "function",
        "function": {
            "name": "market_scan",
            "description": "Scan the full A-share market and return top candidates",
            "parameters": {
                "type": "object",
                "properties": {
                    "top_n": {"type": "integer", "default": 20},
                    "min_pe": {"type": "number"},
                    "max_pe": {"type": "number"},
                    "exclude_sectors": {"type": "array", "items": {"type": "string"}},
                }
            }
        }
    }
]

📈 Sample Output

{
  "code": "002475",
  "name": "立讯精密",
  "price": 71.29,
  "change_pct": 2.55,
  "total_score": 55.0,
  "signal": "WATCH",
  "technical": {
    "score": 55,
    "signals": ["5连阳+5"],
    "consecutive_up_candles": 5
  },
  "capital": {
    "score": 60,
    "signals": ["主力偏强 +28.3万", "外盘偏强+8"]
  },
  "fundamental": {
    "score": 58,
    "pe": 30.17,
    "pb": 5.88
  }
}

🛡️ Risk Management Rules

Built-in anti-chase and risk control rules:

• ❌ No buying stocks with PE > 50 (unless strong catalyst)
• ❌ No chasing stocks up > 5% in a single day
• ❌ No buying stocks that overlap with existing holdings
• ❌ Must have clear stop-loss level before entry
• ✅ Prefer stocks with PE < 30 and positive capital flow
• ✅ Require multi-day confirmation before entry

🙏 Acknowledgments

• OpenClaw — Agent orchestration framework
• AkShare — Financial data API
• Tushare — A-share data provider

License

MIT