Moon Dev's AI Trading Agents System

When to Use This Skill

• Working with Moon Dev's trading agents repository
• Need to understand agent architecture and capabilities
• Running, modifying, or creating trading agents
• Configuring trading system, exchanges, or LLM providers
• Debugging trading operations or agent interactions
• Understanding backtesting with RBI agent
• Setting up new exchanges or strategies

Environment Setup Note

tflow

Quick Start Commands

# Activate your Python environment (conda, venv, or whatever you use)
# Example with conda: conda activate tflow
# Example with venv: source venv/bin/activate
# Use whatever environment manager you prefer

# Run main orchestrator (controls multiple agents)
python src/main.py

# Run individual agent
python src/agents/trading_agent.py
python src/agents/risk_agent.py
python src/agents/rbi_agent.py

# Update requirements after adding packages
pip freeze > requirements.txt

Core Architecture

Directory Structure

src/
├── agents/              # 48+ specialized AI agents (<800 lines each)
├── models/              # LLM provider abstraction (ModelFactory)
├── strategies/          # User-defined trading strategies
├── scripts/             # Standalone utility scripts
├── data/                # Agent outputs, memory, analysis results
├── config.py            # Global configuration
├── main.py              # Main orchestrator loop
├── nice_funcs.py        # Core trading utilities (~1,200 lines)
├── nice_funcs_hl.py     # Hyperliquid-specific functions
├── nice_funcs_extended.py # Extended Exchange functions
└── ezbot.py             # Legacy trading controller

Key Components

• Each agent is standalone executable
• Uses ModelFactory for LLM access
• Stores outputs in src/data/[agent_name]/
• Under 800 lines (split if longer)

• ModelFactory provides unified interface
• Supports: Claude, GPT-4, DeepSeek, Groq, Gemini, Ollama
• Pattern:

  ModelFactory.create_model('anthropic')

• nice_funcs.py

  : Core functions (Solana/BirdEye)

• nice_funcs_hl.py

  : Hyperliquid exchange

• nice_funcs_extended.py

  : Extended Exchange (X10)

• config.py

  : Trading settings, risk limits, agent behavior

• .env

  : API keys and secrets (never expose these)

Agent Categories

Common Workflows

1. Run Single Agent

# Activate your environment first
python src/agents/[agent_name].py

2. Run Main Orchestrator

python src/main.py

ACTIVE_AGENTS

3. Change Exchange

EXCHANGE = "hyperliquid"  # or "birdeye", "extended"

if EXCHANGE == "hyperliquid":
    from src import nice_funcs_hl as nf
elif EXCHANGE == "extended":
    from src import nice_funcs_extended as nf

4. Switch AI Model

src/config.py

AI_MODEL = "claude-3-haiku-20240307"  # Fast, cheap
# AI_MODEL = "claude-3-sonnet-20240229"  # Balanced
# AI_MODEL = "claude-3-opus-20240229"  # Most powerful

from src.models.model_factory import ModelFactory
model = ModelFactory.create_model('deepseek')  # or 'openai', 'groq', etc.
response = model.generate_response(system_prompt, user_content, temperature, max_tokens)

5. Backtest Strategy (RBI Agent)

python src/agents/rbi_agent.py

Development Rules

CRITICAL Rules

1. Keep files under 800 lines - split into new files if longer
2. NEVER move files - can create new, but no moving without asking
3. Use existing environment - don't create new virtual environments, use the one from initial setup
4. Update requirements.txt after any pip install:

   pip freeze > requirements.txt

5. Use real data only - never synthetic/fake data
6. Minimal error handling - user wants to see errors, not over-engineered try/except
7. Never expose API keys - don't show .env contents

Agent Development Pattern

# 1. Use ModelFactory for LLM
from src.models.model_factory import ModelFactory
model = ModelFactory.create_model('anthropic')

# 2. Store outputs in src/data/
output_dir = "src/data/my_agent/"

# 3. Make independently executable
if __name__ == "__main__":
    # Standalone logic here

# 4. Follow naming: [purpose]_agent.py

# 5. Add to config.py if needed

Backtesting

• Use

  backtesting.py

  library (NOT built-in indicators)
• Use

  pandas_ta

  or

  talib

  for indicators
• Sample data:

  src/data/rbi/BTC-USD-15m.csv

Configuration Files

• MONITORED_TOKENS

  ,

  EXCLUDED_TOKENS

• Position sizing:

  usd_size

  ,

  max_usd_order_size

• Risk:

  CASH_PERCENTAGE

  ,

  MAX_LOSS_USD

  ,

  MAX_GAIN_USD

• Agent:

  SLEEP_BETWEEN_RUNS_MINUTES

  ,

  ACTIVE_AGENTS

• AI:

  AI_MODEL

  ,

  AI_MAX_TOKENS

  ,

  AI_TEMPERATURE

• Trading APIs:

  BIRDEYE_API_KEY

  ,

  MOONDEV_API_KEY

  ,

  COINGECKO_API_KEY

• AI:

  ANTHROPIC_KEY

  ,

  OPENAI_KEY

  ,

  DEEPSEEK_KEY

  ,

  GROQ_API_KEY

  ,

  GEMINI_KEY

• Blockchain:

  SOLANA_PRIVATE_KEY

  ,

  HYPER_LIQUID_ETH_PRIVATE_KEY

  ,

  RPC_ENDPOINT

• Extended:

  X10_API_KEY

  ,

  X10_PRIVATE_KEY

  ,

  X10_PUBLIC_KEY

  ,

  X10_VAULT_ID

Exchange Support

nice_funcs_hl.py

• EVM-compatible perpetuals DEX
• Functions:

  market_buy()

  ,

  market_sell()

  ,

  get_position()

  ,

  close_position()

• Leverage up to 50x

nice_funcs.py

• Solana spot token data and trading
• Functions:

  token_overview()

  ,

  token_price()

  ,

  get_ohlcv_data()

• Real-time market data for 15,000+ tokens

nice_funcs_extended.py

• StarkNet-based perpetuals (X10)
• Auto symbol conversion (BTC → BTC-USD)
• Leverage up to 20x
• Functions match Hyperliquid API for compatibility

Data Flow Pattern

Config/Input → Agent Init → API Data Fetch → Data Parsing →
LLM Analysis (via ModelFactory) → Decision Output →
Result Storage (CSV/JSON in src/data/) → Optional Trade Execution

Common Tasks

# Make sure your environment is activated first
pip install package-name
pip freeze > requirements.txt

from src.nice_funcs import token_overview, get_ohlcv_data, token_price

overview = token_overview(token_address)
ohlcv = get_ohlcv_data(token_address, timeframe='1H', days_back=3)
price = token_price(token_address)

from src import nice_funcs_hl as nf
nf.market_buy("BTC", usd_amount=100, leverage=10)
position = nf.get_position("BTC")
nf.close_position("BTC")

from src import nice_funcs_extended as nf
nf.market_buy("BTC", usd_amount=100, leverage=15)
position = nf.get_position("BTC")
nf.close_position("BTC")

Git Operations

• dc55e90: websearch agent
• 921ead6: websearch_agent launched and rbi agent updated
• 6bb55c2: backtest dash

• .env_example
• src/agents/swarm_agent.py
• src/agents/trading_agent.py
• src/data/ohlcv_collector.py

Documentation

• CLAUDE.md

  : Project overview and development guidelines

• hyperliquid.md

  ,

  hyperliquid_setup.md

  : Hyperliquid exchange

• extended_exchange.md

  : Extended Exchange (X10) setup

• rbi_agent.md

  : Research-Based Inference agent

• websearch_agent.md

  : Web search capabilities

• swarm_agent.md

  : Multi-agent coordination

• [agent_name].md

  : Individual agent docs

• Root

  README.md

  : Project overview

• src/models/README.md

  : LLM provider guide

Risk Management

• Risk Agent runs FIRST before any trading decisions
• Circuit breakers:

  MAX_LOSS_USD

  ,

  MINIMUM_BALANCE_USD

• AI confirmation for position-closing (configurable)
• Default loop: every 15 minutes (

  SLEEP_BETWEEN_RUNS_MINUTES

  )

Philosophy

• No guarantees of profitability
• Open source and free
• YouTube-driven development
• Community-supported via Discord
• No official token (avoid scams)

Additional Resources