AI Index Trading Mode

Overview

The AI Index Trading Mode is an advanced trading mode that inherits from IndexTradingMode and uses external AI agents to dynamically generate portfolio distributions based on strategy evaluation data. This mode combines the robust rebalancing infrastructure of index trading with AI-driven decision making for optimal asset allocation, where AI logic is handled by separate agents rather than embedded in the mode.

Key Features

• Agent-Driven Allocations: Uses external AI agents to analyze strategy signals and generate optimal portfolio weights
• Strategy Integration: Collects data from TA, Social, and Real-time evaluator signals on matrix callbacks
• Decoupled AI: AI processing happens in separate agents, allowing for scalability and modularity
• Detailed Instructions: Agents provide actionable rebalance instructions with explanations
• Inherits IndexTradingMode: Full rebalancing, order management, and portfolio optimization capabilities
• Configurable Parameters: Model selection, temperature, token limits for agents

Configuration

AI Configuration Parameters

Parameter	Type	Default	Description
model	string	"gpt-4"	GPT model used by agents (gpt-3.5-turbo, gpt-4, gpt-4-turbo)
temperature	float	0.3	AI creativity for agents (0.0 = deterministic, 1.0 = very creative)
max_tokens	int	2000	Maximum tokens for agent AI responses

Index Trading Parameters (inherited)

Parameter	Type	Default	Description
refresh_interval	int	1	Days between rebalance checks
rebalance_trigger_min_percent	int	5	Minimum % deviation before rebalance

How It Works

1. Strategy Signal Collection

The AI Index Trading Mode monitors matrix callbacks from strategy evaluators:

• Technical Analysis (TA) signals
• Social sentiment signals
• Real-time evaluation signals

2. Data Submission

When strategies update, the mode:

1. Collects current strategy evaluation data
2. Submits neutral evaluations with {"strategy_data": data} to trigger agent processing
3. Agents listen for these submissions and process the data

3. Agent Processing

External AI agents:

1. Receive strategy data from mode submissions
2. Analyze signals using configured GPT models
3. Generate rebalance instructions
4. Provide instructions back to the mode

4. Instruction Application

The mode receives agent-generated instructions:

• Applies distribution changes via ai_index_distribution.apply_ai_instructions
• Validates instructions for consistency
• Executes rebalancing using IndexTradingMode logic

Agent Architecture

Data Flow

Strategy Callbacks → Mode Producer → Submit {"strategy_data": data}
                        ↓
                Agents Process → Generate {"ai_instructions": instructions}
                        ↓
Mode Consumer → Apply Instructions → Rebalance Portfolio

Instruction Format

Agents provide instructions as lists of actions:

[
  {"action": "reduce_exposure", "symbol": "BTC", "amount": 10, "explanation": "Overbought signals"},
  {"action": "increase_exposure", "symbol": "ETH", "amount": 15, "explanation": "Strong momentum"}
]

Agent Responsibilities

• Signal Analysis: Process strategy data with AI models
• Instruction Generation: Create validated rebalance actions
• Error Handling: Handle API failures and provide safe instructions

Usage Examples

Basic Configuration

{
  "trading_mode": "AIIndexTradingMode",
  "model": "gpt-4",
  "temperature": 0.3,
  "max_tokens": 2000,
  "refresh_interval": 1,
  "rebalance_trigger_min_percent": 5
}

Conservative Configuration

{
  "trading_mode": "AIIndexTradingMode",
  "model": "gpt-3.5-turbo",
  "temperature": 0.1,
  "max_tokens": 1500,
  "refresh_interval": 2
}

Testing

Use the tentacles-agent tools to test:

# Test with strategy evaluators
python tentacle_trading_mode_tester.py --mode AIIndexTradingMode --evaluators strategy_evaluators.json --symbol BTC/USDT --duration 120

# Test basic functionality
python tentacle_configuration_tester.py --config ai_index_config.json --validate

Dependencies

• AI Agents: Required for instruction generation
• Strategy Evaluators: Any evaluators providing TA/Social/Real-time signals
• IndexTradingMode: Inherited rebalancing functionality

Troubleshooting

Common Issues

No rebalancing occurs

• Verify strategy evaluators are active and triggering callbacks
• Check that agents are running and processing submissions
• Ensure sufficient portfolio balance for rebalancing

Agents not responding

• Check agent configuration and connectivity
• Verify AI service availability
• Review agent logs for processing errors

Logs to Check

• Strategy data submissions: "Submitting strategy data for AI processing"
• Instruction application: "Applied AI instructions: {...}"
• Validation errors: "Invalid AI instructions received"

Future Enhancements

• Agent Marketplace: Multiple competing AI agents
• Historical Learning: Agents incorporate backtesting results
• Real-time Adaptation: Agents adjust to live market conditions
• Multi-Asset Correlation: Advanced portfolio optimization
• Custom Agent Development: Framework for user-defined agents