Trade Agent

Modular AI trading agent with plugin system.
LLM analyzes market + news + indicators → decision → execution.
Every component swappable. No framework. No lock-in.

---

🇮🇩 Bahasa Indonesia

Apa ini?

Bot trading AI modular dengan sistem plugin. LLM analisa data market + news + indikator teknikal → decision → eksekusi trade.

Setiap komponen bisa diganti via plugin. Nggak ada framework, nggak ada lock-in.

Quick Start

git clone https://github.com/mocasus/trade-agent.git
cd trade-agent
pip install -e ".[all]"
cp config.example.yaml config.yaml

echo "AI_API_KEY=sk-xxx" >> .env
echo "BINANCE_API_KEY=xxx" >> .env
echo "BINANCE_API_SECRET=xxx" >> .env

python -m trade_agent --config config.yaml  # Paper mode default

Plugin System

6 slot yang bisa diganti:

• data_source: ccxt | csv
• strategy: llm | rule | partial_exit | rule_builder
• exchange: ccxt | paper | bybit_futures
• notifier: telegram | discord | webhook | console
• sentiment: rss_llm | none
• risk_profile: conservative | moderate | aggressive | trailing_stop | kelly_sizing

Safety

• Paper mode ON default
• Paper mode forced 24h pertama
• Kill switch: Telegram /stop
• Daily loss limit hard stop
• Audit trail: semua decision di-log

---

✨ Features

• 🧩 Plugin system — 7 swappable slots, 21 built-in plugins
• 🤖 LLM strategy — OpenAI-compatible API analyzes market context → structured decision
• 📊 Technical indicators — RSI, MACD, EMA, Bollinger Bands, ATR
• 🔄 Trailing stop-loss — follows price up, never goes down
• 📐 Kelly criterion — optimal fractional position sizing
• 🎯 Partial exits — scaled exits at profit levels (25%/25%/50%)
• 📋 Rule builder — config-only strategy, no code needed
• 📰 News sentiment — RSS + LLM sentiment analysis
• 🔔 Multi-notifier — Telegram, Discord, webhook, console
• 📈 Backtest engine — historical replay + Sharpe/drawdown/win rate
• 🖥️ Web dashboard — dark theme, real-time portfolio monitoring
• 🐳 Docker compose — multi-instance deployment
• 🛡️ Safety-first — paper mode default, kill switch, daily loss limit, audit trail

Quick Start

git clone https://github.com/mocasus/trade-agent.git
cd trade-agent
pip install -e ".[all]"
cp config.example.yaml config.yaml  # Edit before running

echo "AI_API_KEY=sk-xxx" >> .env
echo "BINANCE_API_KEY=xxx" >> .env

python -m trade_agent --config config.yaml  # Paper mode by default

Architecture

┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐
│DataSource│ │ Strategy │ │ Exchange │ │ Notifier │ │Sentiment │
│ ccxt/csv │ │llm/rule  │ │paper/real│ │ TG/Disc  │ │ rss_llm  │
└──────────┘ └──────────┘ └──────────┘ └──────────┘ └──────────┘
      │            │            │
      ▼            ▼            ▼
┌─────────┐  ┌──────────┐  ┌─────────┐
│Data Layer│─▶│ Strategy │─▶│Execution│
└─────────┘  └──────────┘  └─────────┘
                                 │
              ┌─────────┐       ▼
              │ Storage │  ┌─────────┐
              │ SQLite  │  │Notifier │
              └─────────┘  └─────────┘

Core loop only knows interfaces. Swap any plugin without touching core.

Plugin System

Slot	Interface	Built-in Plugins
data_source	DataSourceInterface	ccxt, csv
strategy	StrategyInterface	llm, rule, partial_exit, rule_builder
exchange	ExchangeInterface	ccxt, paper, bybit_futures
notifier	NotifierInterface	telegram, discord, webhook, console
sentiment	SentimentInterface	rss_llm, none
risk_profile	RiskProfileInterface	conservative, moderate, aggressive, trailing_stop, kelly_sizing
indicators	IndicatorPluginInterface	ta (RSI, MACD, EMA, BB, ATR)

Custom plugin — implement interface → place in ~/.trade-agent/plugins/<slot>/

# ~/.trade-agent/plugins/strategy/my_quant.py
from trade_agent.interfaces import StrategyInterface
from trade_agent.models import Decision, MarketContext, Action

class MyQuantStrategy(StrategyInterface):
    def analyze(self, context: MarketContext) -> Decision:
        return Decision(action=Action.BUY, symbol=context.symbol,
                        confidence=80, reasoning="...")

def register():
    return {"name": "my_quant", "class": MyQuantStrategy}

Config: plugins.strategy: "my_quant"

Advanced Features

Trailing Stop-Loss

risk:
  method: trailing_stop
  trailing_stop:
    activation_threshold: 0.03  # Activate after 3% profit
    trail_percent: 0.015        # Trail 1.5% below peak

Kelly Criterion Sizing

risk:
  position_sizing:
    method: kelly
    kelly_fraction: 0.5  # Half-Kelly (recommended)

Bybit Futures

plugins:
  exchange: bybit_futures
exchange:
  bybit_futures:
    leverage: 3
    margin_mode: cross

Partial Exit

strategy:
  partial_exit:
    levels:
      - profit_pct: 5,  exit_pct: 25
      - profit_pct: 10, exit_pct: 25
      - profit_pct: 15, exit_pct: 50

Rule Builder

strategy:
  rule_builder:
    rules:
      - name: "RSI oversold buy"
        conditions: [{indicator: rsi, operator: "<", value: 30}]
        action: buy, confidence: 75
      - name: "MACD sell"
        conditions: [{indicator: macd_histogram, operator: "<", value: 0}]
        action: sell, confidence: 70

Web Dashboard

python -m trade_agent.dashboard --config config.yaml  # :8080

Risk Profiles

	Conservative	Moderate	Aggressive	Trailing Stop	Kelly
Max position	2%	8%	15%	10%	Dynamic
Confidence floor	70	65	55	60	60
Stop-loss	ATR×1.0	ATR×1.5	ATR×2.0	Trail 1.5%	ATR×1.5
R:R ratio	1.5	2.0	1.5	Variable	2.0
Daily loss limit	2%	5%	10%	5%	5%

🛡️ Safety

• Paper mode ON by default — won't trade real money
• Paper mode forced 24h — can't accidentally go live
• Kill switch — Telegram /stop halts immediately
• Daily loss limit — hard stop, AI can't override
• Max position size — AI can't exceed configured %
• Audit trail — every decision logged with reasoning
• Auto-stop on error — halts on unhandled exceptions
• Max API failures — stops after N consecutive failures

Backtesting

python -m trade_agent.backtest --config config.yaml

Outputs: total return, win rate, Sharpe ratio, max drawdown.

📋 Full test results → — 39 tests, 7 suites, all passing.

Deploy

systemd

sudo cp trade-agent.service /etc/systemd/system/
sudo systemctl enable --now trade-agent

Docker

docker-compose up -d

Project Structure

trade_agent/
├── agent.py              # Main loop
├── interfaces.py         # 6 abstract interfaces
├── plugin_loader.py      # Plugin discovery + loading
├── config.py             # YAML config + env var resolution
├── models.py             # Data models
├── storage.py            # SQLite persistence
├── risk.py               # Built-in risk profiles
├── backtest.py           # Historical replay engine
├── dashboard/            # Web dashboard
│   ├── server.py
│   └── templates/
└── plugins/              # Built-in plugins
    ├── data_source/      # ccxt, csv
    ├── strategy/         # llm, rule, partial_exit, rule_builder
    ├── exchange/         # ccxt, paper, bybit_futures
    ├── notifier/         # telegram, discord, webhook, console
    ├── sentiment/        # rss_llm, none
    ├── risk_profile/     # conservative, moderate, aggressive, trailing_stop, kelly_sizing
    └── indicators/       # ta (RSI, MACD, EMA, BB, ATR)

License

MIT

---

Made with

Supported Exchanges

Integrations

Status

---

_{v1.0.0 · 2026 · Built by @mocasus}