AI Trading Crew

AI Trading Crew

USE FOR:

• "50-agent trading crew simulation"
• "AutoGen multi-agent stock analysis"
• "devil's advocate + risk veto trading system"
• "US stock consensus trading agent"
• "multi-team agent debate for trading decisions"
• "ChromaDB RAG for trading knowledge" tags: [AutoGen, multi-agent, trading, US-stocks, Alpaca, Polygon, ChromaDB, RAG, risk-veto, paper-trading] kind: framework category: quant-ml-trading

What Is AI Trading Crew?

50-agent AutoGen system simulating a collaborative trading firm for US equities.

• Repo: https://github.com/omer475/ai-trading-crew
• Framework: AutoGen (Microsoft multi-agent)
• LLM: OpenAI
• Broker: Alpaca (paper trading)
• Data: Polygon.io real-time market data
• Memory: ChromaDB RAG knowledge base

Agent Architecture: 8 Teams + Head Coach

                    Head Coach (Supervisor)
                          ↑ synthesis
    ┌──────────┬──────────┼──────────┬──────────┐
    │          │          │          │          │
Technical  Fundamental  Macro    Sentiment   Quant
(7 agents) (7 agents) (6 agents) (6 agents) (6 agents)
    │          │          │          │          │
    └──────────┴──────────┼──────────┴──────────┘
                          │
                    Risk Management (5 agents) ← VETO POWER
                          │ approved?
                    Execution & Ops (5 agents)
                    Strategy & Special (8 agents)
                          │
                    Devil's Advocate ← contrarian challenge
                          │
                    Final Decision + Order

Team Responsibilities

Trading Workflow

1. Input: python main.py --symbol AAPL

2. Teams debate internally via AutoGen GroupChat
   → Each team reaches internal consensus

3. Team leaders report to Head Coach
   → Cross-team synthesis

4. Risk Management review
   → Can VETO any trade (overrides Head Coach)

5. Devil's Advocate challenges recommendation
   → Forces bull/bear stress test

6. Head Coach final decision

7. Human approval gate (configurable)

8. Execution team submits order to Alpaca

Installation

git clone https://github.com/omer475/ai-trading-crew
cd agents
pip install -r requirements.txt
cp .env.example .env

.env keys required:

OPENAI_API_KEY="sk-..."
ALPACA_API_KEY="..."
ALPACA_SECRET_KEY="..."
POLYGON_API_KEY="..."

Usage

# Full 50-agent analysis
python main.py --symbol AAPL

# Quick 5-agent test mode
python main.py --symbol NVDA --test

# Output: consensus decision + rationale + risk assessment + order

Key Design Patterns

AutoGen GroupChat per Team

# Each team runs internal debate
groupchat = autogen.GroupChat(
    agents=[tech_agent_1, tech_agent_2, ..., tech_agent_7],
    messages=[],
    max_round=5
)
manager = autogen.GroupChatManager(groupchat=groupchat, llm_config=llm_config)

Risk Veto Pattern

class RiskManager(autogen.AssistantAgent):
    def check_veto(self, proposal: dict) -> bool:
        if proposal["position_size"] > self.max_risk:
            return True   # VETO
        if proposal["volatility"] > self.vol_threshold:
            return True   # VETO
        return False      # Approved

ChromaDB RAG Knowledge Base

import chromadb
client = chromadb.Client()
collection = client.get_or_create_collection("trading_knowledge")

# Query before analysis
results = collection.query(
    query_texts=["AAPL earnings history semiconductor cycle"],
    n_results=5
)

Unique Features vs Other Trading Agent Frameworks