# Trading Agent Service Design

**Version**: 1.0
**Date**: 2025-10-16
**Author**: Agent 11.10
**Status**: Design Phase

---

## Executive Summary

The **Trading Agent Service** is a new microservice that orchestrates trading decisions by managing universe selection, asset selection, portfolio allocation, and strategy coordination. It **drives** the Trading Service by generating and submitting orders based on ML predictions, market conditions, and risk constraints.

**Key Principle**: The Trading Agent Service is the **decision-making brain** while the Trading Service remains the **execution engine**.

---

## Architecture Overview

### Current State
```
API Gateway (50051)
    ↓
Trading Service (50052) ← monolithic decision + execution
Backtesting Service (50053)
ML Training Service (50054)
```

### Target State
```
API Gateway (50051)
    ↓
Trading Agent Service (50055) ← NEW: decision-making orchestration
    ↓ (drives via gRPC calls)
Trading Service (50052) ← execution only
    ↓
ML Training Service (50054) ← provides ML predictions
Backtesting Service (50053) ← simulates Trading Agent behavior
```

---

## Service Responsibilities

### Trading Agent Service (NEW - Port 50055)

**Core Responsibilities**:
1. **Universe Selection**: Determine which markets/exchanges to trade (ES.FUT, NQ.FUT, etc.)
2. **Asset Selection**: Choose specific instruments within universe based on ML signals
3. **Portfolio Allocation**: Optimize capital allocation across selected assets
4. **Risk Management Coordination**: Enforce portfolio-level risk limits
5. **Strategy Orchestration**: Coordinate multiple trading strategies (ML ensemble, mean reversion, etc.)
6. **Order Generation**: Create orders based on allocation decisions
7. **ML Integration**: Query ML Training Service for predictions
8. **Performance Monitoring**: Track agent performance vs. benchmarks

**What it DOES**:
- ✅ Decides WHAT to trade and WHEN
- ✅ Calculates position sizes and allocations
- ✅ Generates order instructions
- ✅ Monitors overall portfolio health
- ✅ Adapts to market regimes

**What it DOES NOT do**:
- ❌ Execute orders (Trading Service responsibility)
- ❌ Manage individual order lifecycle (Trading Service)
- ❌ Track fills and positions (Trading Service)
- ❌ Stream market data (Trading Service)
- ❌ Train ML models (ML Training Service)

### Trading Service (Port 50052)

**Responsibilities (UNCHANGED)**:
- Order execution and lifecycle management
- Position tracking and PnL calculation
- Market data streaming
- Execution quality monitoring
- Paper trading simulation
- ML order submission (enhanced with agent integration)

**New Integration**:
- Receives order instructions from Trading Agent Service
- Reports execution status back to Trading Agent Service
- Provides position snapshots for allocation decisions

---

## gRPC API Design

### Proto Definition

**File**: `services/trading_agent_service/proto/trading_agent.proto`

```protobuf
syntax = "proto3";

package trading_agent;

// Trading Agent Service orchestrates trading decisions across universe selection,
// asset selection, portfolio allocation, and strategy coordination.
service TradingAgentService {
  // Universe Management
  // Select tradable universe based on liquidity, volatility, and ML signals
  rpc SelectUniverse(SelectUniverseRequest) returns (SelectUniverseResponse);

  // Get current trading universe configuration
  rpc GetUniverse(GetUniverseRequest) returns (GetUniverseResponse);

  // Update universe selection criteria
  rpc UpdateUniverseCriteria(UpdateUniverseCriteriaRequest) returns (UpdateUniverseCriteriaResponse);

  // Asset Selection
  // Select specific assets to trade within universe
  rpc SelectAssets(SelectAssetsRequest) returns (SelectAssetsResponse);

  // Get current asset selection with scores
  rpc GetSelectedAssets(GetSelectedAssetsRequest) returns (GetSelectedAssetsResponse);

  // Portfolio Allocation
  // Allocate capital across selected assets
  rpc AllocatePortfolio(AllocatePortfolioRequest) returns (AllocatePortfolioResponse);

  // Get current portfolio allocation
  rpc GetAllocation(GetAllocationRequest) returns (GetAllocationResponse);

  // Rebalance portfolio based on target allocation
  rpc RebalancePortfolio(RebalancePortfolioRequest) returns (RebalancePortfolioResponse);

  // Order Generation
  // Generate orders based on allocation and ML signals
  rpc GenerateOrders(GenerateOrdersRequest) returns (GenerateOrdersResponse);

  // Submit generated orders to Trading Service
  rpc SubmitAgentOrders(SubmitAgentOrdersRequest) returns (SubmitAgentOrdersResponse);

  // Strategy Coordination
  // Register a trading strategy with the agent
  rpc RegisterStrategy(RegisterStrategyRequest) returns (RegisterStrategyResponse);

  // Get list of active strategies
  rpc ListStrategies(ListStrategiesRequest) returns (ListStrategiesResponse);

  // Enable/disable a strategy
  rpc UpdateStrategyStatus(UpdateStrategyStatusRequest) returns (UpdateStrategyStatusResponse);

  // Agent Monitoring
  // Get comprehensive agent status and performance
  rpc GetAgentStatus(GetAgentStatusRequest) returns (GetAgentStatusResponse);

  // Stream real-time agent decisions and actions
  rpc StreamAgentActivity(StreamAgentActivityRequest) returns (stream AgentActivityEvent);

  // Get agent performance metrics
  rpc GetAgentPerformance(GetAgentPerformanceRequest) returns (GetAgentPerformanceResponse);

  // Service Health
  rpc HealthCheck(HealthCheckRequest) returns (HealthCheckResponse);
}

// Universe Selection Messages

message SelectUniverseRequest {
  UniverseCriteria criteria = 1;       // Selection criteria
  optional uint32 max_instruments = 2; // Maximum instruments in universe
  bool force_refresh = 3;              // Force recalculation
}

message SelectUniverseResponse {
  repeated Instrument instruments = 1; // Selected instruments
  UniverseMetrics metrics = 2;         // Universe quality metrics
  int64 timestamp = 3;                 // Selection timestamp (nanoseconds)
  string universe_id = 4;              // Unique universe identifier
}

message GetUniverseRequest {
  optional string universe_id = 1;     // Get specific universe, or current if not specified
}

message GetUniverseResponse {
  string universe_id = 1;
  repeated Instrument instruments = 2;
  UniverseCriteria criteria = 3;
  UniverseMetrics metrics = 4;
  int64 created_at = 5;                // Unix timestamp (nanoseconds)
  int64 updated_at = 6;
}

message UpdateUniverseCriteriaRequest {
  UniverseCriteria criteria = 1;
}

message UpdateUniverseCriteriaResponse {
  bool success = 1;
  string message = 2;
  string universe_id = 3;              // New universe ID after update
}

// Asset Selection Messages

message SelectAssetsRequest {
  string universe_id = 1;              // Universe to select from
  AssetSelectionCriteria criteria = 2; // Selection criteria
  uint32 max_assets = 3;               // Maximum assets to select
}

message SelectAssetsResponse {
  repeated AssetScore assets = 1;      // Selected assets with scores
  SelectionMetrics metrics = 2;        // Selection quality metrics
  int64 timestamp = 3;
}

message GetSelectedAssetsRequest {
  optional string universe_id = 1;
}

message GetSelectedAssetsResponse {
  repeated AssetScore assets = 1;
  SelectionMetrics metrics = 2;
  int64 timestamp = 3;
}

// Portfolio Allocation Messages

message AllocatePortfolioRequest {
  repeated AssetScore assets = 1;      // Assets to allocate across
  AllocationStrategy strategy = 2;     // Allocation algorithm
  RiskConstraints risk_constraints = 3; // Risk limits
  double total_capital = 4;            // Total capital to allocate
}

message AllocatePortfolioResponse {
  repeated AssetAllocation allocations = 1; // Allocation per asset
  AllocationMetrics metrics = 2;       // Allocation quality metrics
  int64 timestamp = 3;
  string allocation_id = 4;
}

message GetAllocationRequest {
  optional string allocation_id = 1;   // Get specific allocation, or current if not specified
}

message GetAllocationResponse {
  string allocation_id = 1;
  repeated AssetAllocation allocations = 2;
  AllocationMetrics metrics = 3;
  int64 created_at = 4;
  double total_capital = 5;
}

message RebalancePortfolioRequest {
  string allocation_id = 1;            // Target allocation
  double rebalance_threshold = 2;      // Min deviation to trigger rebalance (%)
  bool force_rebalance = 3;            // Force rebalance regardless of threshold
}

message RebalancePortfolioResponse {
  repeated RebalanceAction actions = 1; // Required rebalancing actions
  RebalanceMetrics metrics = 2;
  bool rebalance_required = 3;
  int64 timestamp = 4;
}

// Order Generation Messages

message GenerateOrdersRequest {
  string allocation_id = 1;            // Target allocation
  repeated MLSignal ml_signals = 2;    // ML predictions for timing
  OrderGenerationStrategy strategy = 3; // Order generation algorithm
}

message GenerateOrdersResponse {
  repeated GeneratedOrder orders = 1;  // Generated order instructions
  OrderGenerationMetrics metrics = 2;
  int64 timestamp = 3;
  string order_batch_id = 4;
}

message SubmitAgentOrdersRequest {
  string order_batch_id = 1;           // Batch ID from GenerateOrders
  repeated GeneratedOrder orders = 2;  // Orders to submit
  bool dry_run = 3;                    // Test without actual submission
}

message SubmitAgentOrdersResponse {
  repeated OrderSubmissionResult results = 1; // Submission results per order
  OrderSubmissionMetrics metrics = 2;
  int64 timestamp = 3;
}

// Strategy Coordination Messages

message RegisterStrategyRequest {
  string strategy_name = 1;            // Unique strategy name
  StrategyType strategy_type = 2;      // Strategy category
  StrategyConfig config = 3;           // Strategy configuration
  bool auto_enable = 4;                // Enable immediately after registration
}

message RegisterStrategyResponse {
  bool success = 1;
  string strategy_id = 2;
  string message = 3;
}

message ListStrategiesRequest {
  optional StrategyStatus status_filter = 1; // Filter by status
}

message ListStrategiesResponse {
  repeated Strategy strategies = 1;
}

message UpdateStrategyStatusRequest {
  string strategy_id = 1;
  StrategyStatus new_status = 2;
  optional string reason = 3;
}

message UpdateStrategyStatusResponse {
  bool success = 1;
  string message = 2;
  Strategy updated_strategy = 3;
}

// Agent Monitoring Messages

message GetAgentStatusRequest {
  bool include_performance = 1;        // Include performance metrics
  bool include_positions = 2;          // Include current positions
}

message GetAgentStatusResponse {
  AgentStatus status = 1;
  optional AgentPerformanceMetrics performance = 2;
  optional PositionSummary positions = 3;
  int64 timestamp = 4;
}

message StreamAgentActivityRequest {
  repeated ActivityType activity_types = 1; // Filter by activity type
}

message AgentActivityEvent {
  ActivityType activity_type = 1;
  oneof event {
    UniverseSelectionEvent universe_event = 2;
    AssetSelectionEvent asset_event = 3;
    AllocationEvent allocation_event = 4;
    OrderGenerationEvent order_event = 5;
    StrategyEvent strategy_event = 6;
  }
  int64 timestamp = 7;
}

message GetAgentPerformanceRequest {
  optional int64 start_time = 1;       // Performance window start (nanoseconds)
  optional int64 end_time = 2;         // Performance window end (nanoseconds)
  bool include_strategy_breakdown = 3; // Include per-strategy performance
}

message GetAgentPerformanceResponse {
  AgentPerformanceMetrics metrics = 1;
  repeated StrategyPerformance strategy_performance = 2;
  int64 timestamp = 3;
}

message HealthCheckRequest {}

message HealthCheckResponse {
  bool healthy = 1;
  string message = 2;
  map<string, string> details = 3;
}

// Data Structures

message Instrument {
  string symbol = 1;                   // Trading symbol (ES.FUT, NQ.FUT)
  string exchange = 2;                 // Exchange identifier
  InstrumentType instrument_type = 3;  // Futures, equity, FX, etc.
  double liquidity_score = 4;          // Liquidity rating (0.0-1.0)
  double volatility = 5;               // Annualized volatility
  double ml_signal_strength = 6;       // ML prediction confidence
  map<string, string> metadata = 7;
}

message UniverseCriteria {
  double min_liquidity_score = 1;      // Minimum liquidity threshold
  double min_volatility = 2;           // Minimum volatility
  double max_volatility = 3;           // Maximum volatility
  repeated InstrumentType allowed_types = 4;
  repeated string exchanges = 5;       // Allowed exchanges
  double min_ml_confidence = 6;        // Minimum ML signal confidence
}

message UniverseMetrics {
  uint32 total_instruments = 1;
  double avg_liquidity_score = 2;
  double avg_volatility = 3;
  double portfolio_diversification = 4; // 0.0-1.0
}

message AssetSelectionCriteria {
  double min_ml_signal_strength = 1;   // Minimum ML confidence
  double min_sharpe_ratio = 2;         // Minimum risk-adjusted return
  SelectionMode mode = 3;              // Top-N, threshold-based, etc.
}

message AssetScore {
  string symbol = 1;
  double ml_score = 2;                 // ML model prediction score
  double momentum_score = 3;           // Momentum factor score
  double value_score = 4;              // Value factor score
  double quality_score = 5;            // Quality factor score
  double composite_score = 6;          // Final weighted score
  map<string, double> model_scores = 7; // Per-model scores (DQN, MAMBA2, etc.)
}

message SelectionMetrics {
  uint32 assets_evaluated = 1;
  uint32 assets_selected = 2;
  double avg_composite_score = 3;
  double min_score = 4;
  double max_score = 5;
}

message AllocationStrategy {
  AllocationType allocation_type = 1;  // Equal-weight, risk-parity, etc.
  map<string, double> parameters = 2;  // Strategy-specific parameters
}

message RiskConstraints {
  double max_position_size_pct = 1;    // Max % of portfolio per position
  double max_sector_exposure_pct = 2;  // Max % per sector
  double max_volatility = 3;           // Portfolio volatility limit
  double max_var_95 = 4;               // Value at Risk (95%)
  double max_leverage = 5;             // Maximum leverage ratio
}

message AssetAllocation {
  string symbol = 1;
  double target_weight = 2;            // Target allocation weight (0.0-1.0)
  double target_capital = 3;           // Target capital in USD
  double target_quantity = 4;          // Target position size
  double current_weight = 5;           // Current allocation weight
  double current_quantity = 6;         // Current position size
  double rebalance_delta = 7;          // Required change
}

message AllocationMetrics {
  double total_weight = 1;             // Should be ~1.0
  double portfolio_volatility = 2;     // Expected portfolio volatility
  double portfolio_sharpe = 3;         // Expected Sharpe ratio
  double var_95 = 4;                   // Portfolio VaR (95%)
  double max_drawdown_estimate = 5;    // Expected max drawdown
}

message RebalanceAction {
  string symbol = 1;
  double current_quantity = 2;
  double target_quantity = 3;
  double delta_quantity = 4;           // Positive = buy, negative = sell
  RebalanceReason reason = 5;
}

message RebalanceMetrics {
  uint32 total_rebalance_actions = 1;
  double total_turnover = 2;           // Total capital moved (USD)
  double estimated_cost = 3;           // Estimated transaction costs
}

message MLSignal {
  string symbol = 1;
  string model_name = 2;               // DQN, MAMBA2, PPO, TFT
  double signal_strength = 3;          // -1.0 to 1.0 (short to long)
  double confidence = 4;               // 0.0 to 1.0
  string predicted_action = 5;         // BUY, SELL, HOLD
  int64 timestamp = 6;
}

message OrderGenerationStrategy {
  OrderGenerationMode mode = 1;
  double slippage_tolerance = 2;       // Max acceptable slippage (%)
  bool use_limit_orders = 3;           // Use limit orders vs market
  double limit_price_offset = 4;       // Offset from mid price (%)
}

message GeneratedOrder {
  string symbol = 1;
  OrderSide side = 2;                  // BUY or SELL
  double quantity = 3;
  OrderType order_type = 4;            // MARKET, LIMIT, etc.
  optional double price = 5;           // Limit price if applicable
  string rationale = 6;                // Why this order was generated
  map<string, string> metadata = 7;
}

message OrderGenerationMetrics {
  uint32 orders_generated = 1;
  double total_notional = 2;           // Total order value (USD)
  double avg_order_size = 3;
}

message OrderSubmissionResult {
  string symbol = 1;
  bool success = 2;
  optional string order_id = 3;        // From Trading Service
  optional string error_message = 4;
}

message OrderSubmissionMetrics {
  uint32 orders_submitted = 1;
  uint32 orders_accepted = 2;
  uint32 orders_rejected = 3;
  double acceptance_rate = 4;
}

message Strategy {
  string strategy_id = 1;
  string strategy_name = 2;
  StrategyType strategy_type = 3;
  StrategyStatus status = 4;
  StrategyConfig config = 5;
  StrategyPerformance performance = 6;
  int64 created_at = 7;
  int64 updated_at = 8;
}

message StrategyConfig {
  map<string, string> parameters = 1;  // Strategy-specific parameters
  repeated string target_symbols = 2;  // Symbols this strategy trades
  double max_capital_pct = 3;          // Max % of portfolio for this strategy
}

message StrategyPerformance {
  string strategy_id = 1;
  double total_pnl = 2;
  double sharpe_ratio = 3;
  double win_rate = 4;
  uint32 total_trades = 5;
  int64 period_start = 6;
  int64 period_end = 7;
}

message AgentStatus {
  AgentState state = 1;
  string current_universe_id = 2;
  uint32 active_strategies = 3;
  uint32 selected_assets = 4;
  double portfolio_utilization = 5;    // % of capital deployed
  int64 last_action_timestamp = 6;
}

message AgentPerformanceMetrics {
  double total_pnl = 1;
  double sharpe_ratio = 2;
  double max_drawdown = 3;
  double win_rate = 4;
  uint32 total_trades = 5;
  double avg_trade_pnl = 6;
  double portfolio_turnover = 7;       // Annualized
  int64 period_start = 8;
  int64 period_end = 9;
}

message PositionSummary {
  repeated Position positions = 1;
  double total_equity = 2;
  double total_exposure = 3;
  double leverage_ratio = 4;
}

message Position {
  string symbol = 1;
  double quantity = 2;
  double average_price = 3;
  double market_value = 4;
  double unrealized_pnl = 5;
  double weight = 6;                   // % of portfolio
}

message UniverseSelectionEvent {
  string universe_id = 1;
  repeated string added_symbols = 2;
  repeated string removed_symbols = 3;
  UniverseMetrics metrics = 4;
}

message AssetSelectionEvent {
  repeated AssetScore selected_assets = 1;
  SelectionMetrics metrics = 2;
}

message AllocationEvent {
  string allocation_id = 1;
  repeated AssetAllocation allocations = 2;
  AllocationMetrics metrics = 3;
}

message OrderGenerationEvent {
  string order_batch_id = 1;
  repeated GeneratedOrder orders = 2;
  OrderGenerationMetrics metrics = 3;
}

message StrategyEvent {
  string strategy_id = 1;
  StrategyEventType event_type = 2;
  string message = 3;
}

// Enums

enum InstrumentType {
  INSTRUMENT_TYPE_UNSPECIFIED = 0;
  INSTRUMENT_TYPE_EQUITY = 1;
  INSTRUMENT_TYPE_FUTURES = 2;
  INSTRUMENT_TYPE_FX = 3;
  INSTRUMENT_TYPE_OPTIONS = 4;
  INSTRUMENT_TYPE_CRYPTO = 5;
}

enum SelectionMode {
  SELECTION_MODE_UNSPECIFIED = 0;
  SELECTION_MODE_TOP_N = 1;            // Select top N by score
  SELECTION_MODE_THRESHOLD = 2;        // Select all above threshold
  SELECTION_MODE_QUANTILE = 3;         // Select top quantile (e.g., top 20%)
}

enum AllocationType {
  ALLOCATION_TYPE_UNSPECIFIED = 0;
  ALLOCATION_TYPE_EQUAL_WEIGHT = 1;    // 1/N allocation
  ALLOCATION_TYPE_RISK_PARITY = 2;     // Equal risk contribution
  ALLOCATION_TYPE_ML_OPTIMIZED = 3;    // ML-based optimization
  ALLOCATION_TYPE_KELLY = 4;           // Kelly criterion
  ALLOCATION_TYPE_MEAN_VARIANCE = 5;   // Mean-variance optimization
}

enum RebalanceReason {
  REBALANCE_REASON_UNSPECIFIED = 0;
  REBALANCE_REASON_DRIFT = 1;          // Allocation drifted from target
  REBALANCE_REASON_UNIVERSE_CHANGE = 2; // Universe updated
  REBALANCE_REASON_RISK_LIMIT = 3;     // Risk limit violation
  REBALANCE_REASON_MANUAL = 4;         // Manual rebalance request
}

enum OrderGenerationMode {
  ORDER_GENERATION_MODE_UNSPECIFIED = 0;
  ORDER_GENERATION_MODE_AGGRESSIVE = 1; // Market orders, immediate execution
  ORDER_GENERATION_MODE_PASSIVE = 2;   // Limit orders, minimize slippage
  ORDER_GENERATION_MODE_ADAPTIVE = 3;  // Adapt based on market conditions
}

enum OrderSide {
  ORDER_SIDE_UNSPECIFIED = 0;
  ORDER_SIDE_BUY = 1;
  ORDER_SIDE_SELL = 2;
}

enum OrderType {
  ORDER_TYPE_UNSPECIFIED = 0;
  ORDER_TYPE_MARKET = 1;
  ORDER_TYPE_LIMIT = 2;
  ORDER_TYPE_STOP = 3;
  ORDER_TYPE_STOP_LIMIT = 4;
}

enum StrategyType {
  STRATEGY_TYPE_UNSPECIFIED = 0;
  STRATEGY_TYPE_ML_ENSEMBLE = 1;       // Ensemble ML predictions
  STRATEGY_TYPE_MEAN_REVERSION = 2;    // Mean reversion
  STRATEGY_TYPE_MOMENTUM = 3;          // Momentum/trend following
  STRATEGY_TYPE_ARBITRAGE = 4;         // Statistical arbitrage
  STRATEGY_TYPE_MARKET_MAKING = 5;     // Market making
}

enum StrategyStatus {
  STRATEGY_STATUS_UNSPECIFIED = 0;
  STRATEGY_STATUS_ENABLED = 1;
  STRATEGY_STATUS_DISABLED = 2;
  STRATEGY_STATUS_PAUSED = 3;
  STRATEGY_STATUS_ERROR = 4;
}

enum AgentState {
  AGENT_STATE_UNSPECIFIED = 0;
  AGENT_STATE_INITIALIZING = 1;
  AGENT_STATE_ACTIVE = 2;
  AGENT_STATE_PAUSED = 3;
  AGENT_STATE_ERROR = 4;
  AGENT_STATE_SHUTDOWN = 5;
}

enum ActivityType {
  ACTIVITY_TYPE_UNSPECIFIED = 0;
  ACTIVITY_TYPE_UNIVERSE_SELECTION = 1;
  ACTIVITY_TYPE_ASSET_SELECTION = 2;
  ACTIVITY_TYPE_ALLOCATION = 3;
  ACTIVITY_TYPE_ORDER_GENERATION = 4;
  ACTIVITY_TYPE_STRATEGY = 5;
}

enum StrategyEventType {
  STRATEGY_EVENT_TYPE_UNSPECIFIED = 0;
  STRATEGY_EVENT_TYPE_REGISTERED = 1;
  STRATEGY_EVENT_TYPE_ENABLED = 2;
  STRATEGY_EVENT_TYPE_DISABLED = 3;
  STRATEGY_EVENT_TYPE_ERROR = 4;
}
```

---

## Data Flow Diagrams

### 1. Universe Selection Flow
```
User/Scheduler → Trading Agent Service
                      ↓
                SelectUniverse(criteria)
                      ↓
    ┌─────────────────┴─────────────────┐
    │ 1. Query market data (liquidity)   │
    │ 2. Calculate volatility metrics    │
    │ 3. Fetch ML signal strengths       │
    │ 4. Apply selection criteria        │
    │ 5. Rank and filter instruments     │
    └─────────────────┬─────────────────┘
                      ↓
        Universe (ES.FUT, NQ.FUT, ZN.FUT, etc.)
                      ↓
            Store in PostgreSQL
```

### 2. Asset Selection & Allocation Flow
```
Trading Agent Service (scheduled job, e.g., every 5 minutes)
                      ↓
        SelectAssets(universe_id, criteria)
                      ↓
    ┌─────────────────┴─────────────────┐
    │ 1. Get instruments from universe   │
    │ 2. Query ML Training Service       │
    │    → GetMLPredictions(symbols)     │
    │ 3. Calculate factor scores         │
    │ 4. Compute composite scores        │
    │ 5. Rank and select top N assets    │
    └─────────────────┬─────────────────┘
                      ↓
        Selected Assets (ES.FUT, NQ.FUT)
                      ↓
        AllocatePortfolio(assets, strategy, risk)
                      ↓
    ┌─────────────────┴─────────────────┐
    │ 1. Get current positions from      │
    │    Trading Service                 │
    │ 2. Calculate target weights        │
    │ 3. Apply risk constraints          │
    │ 4. Compute target quantities       │
    └─────────────────┬─────────────────┘
                      ↓
    Allocation (ES.FUT: 40%, NQ.FUT: 60%)
```

### 3. Order Generation & Submission Flow
```
Trading Agent Service (allocation + ML signals)
                      ↓
        GenerateOrders(allocation_id, ml_signals)
                      ↓
    ┌─────────────────┴─────────────────┐
    │ 1. Calculate delta from current    │
    │ 2. Determine order side (buy/sell) │
    │ 3. Set order type (market/limit)   │
    │ 4. Apply price offsets             │
    │ 5. Add order metadata              │
    └─────────────────┬─────────────────┘
                      ↓
        Generated Orders (BUY ES.FUT 10 @ MARKET)
                      ↓
        SubmitAgentOrders(orders)
                      ↓
    ┌─────────────────┴─────────────────┐
    │ Trading Agent → Trading Service    │
    │ SubmitMLOrder(symbol, features)    │
    └─────────────────┬─────────────────┘
                      ↓
        Trading Service executes orders
                      ↓
        Returns order IDs and status
                      ↓
    Trading Agent logs execution results
```

### 4. Strategy Coordination Flow
```
User → TLI → API Gateway → Trading Agent Service
                                  ↓
                RegisterStrategy(name, type, config)
                                  ↓
        Store strategy in PostgreSQL
                                  ↓
    ┌───────────────────────────┐
    │ Strategy 1: ML Ensemble   │ ← ENABLED
    │ Strategy 2: Mean Reversion│ ← DISABLED
    │ Strategy 3: Momentum      │ ← ENABLED
    └───────────────────────────┘
                                  ↓
        Periodic execution (every 5 min):
                                  ↓
    ┌─────────────────────────────────┐
    │ For each ENABLED strategy:      │
    │ 1. Run strategy logic           │
    │ 2. Generate allocation          │
    │ 3. Submit orders                │
    │ 4. Track performance            │
    └─────────────────────────────────┘
```

---

## Integration Points

### 1. Trading Agent ↔ Trading Service

**Trading Agent calls Trading Service** (Client → Server):

```rust
// Get current positions for allocation decisions
let positions = trading_service_client
    .get_positions(GetPositionsRequest {
        account_id: Some(agent_account_id),
        symbol: None
    })
    .await?;

// Submit generated orders
let ml_order_result = trading_service_client
    .submit_ml_order(MLOrderRequest {
        symbol: "ES.FUT".to_string(),
        account_id: agent_account_id,
        use_ensemble: true,
        features: feature_vector,
        ..Default::default()
    })
    .await?;
```

**Trading Service reports to Trading Agent** (via callback or stream):
- Order fill notifications
- Position updates
- Execution quality metrics

### 2. Trading Agent ↔ ML Training Service

**Trading Agent queries ML predictions**:

```rust
// Get ML predictions for asset selection
let predictions = ml_training_client
    .get_ml_predictions(GetMLPredictionsRequest {
        symbols: vec!["ES.FUT", "NQ.FUT", "ZN.FUT"],
        models: vec!["DQN", "MAMBA2", "PPO", "TFT"],
        timestamp: current_time,
    })
    .await?;

// Use predictions to score assets
for prediction in predictions {
    let score = calculate_composite_score(&prediction);
    asset_scores.push(AssetScore {
        symbol: prediction.symbol,
        ml_score: score,
        model_scores: prediction.model_scores,
        ..Default::default()
    });
}
```

### 3. TLI ↔ Trading Agent Service

**TLI commands** (via API Gateway):

```bash
# Select trading universe
tli agent universe select --min-liquidity 0.7 --max-volatility 0.5

# View current universe
tli agent universe show

# Select assets
tli agent assets select --top-n 5 --min-ml-score 0.6

# Allocate portfolio
tli agent allocate --strategy risk-parity --capital 1000000

# Generate and submit orders
tli agent orders generate --allocation-id abc123
tli agent orders submit --batch-id xyz789

# Register strategy
tli agent strategy register \
    --name "ml_ensemble_v1" \
    --type ML_ENSEMBLE \
    --config config.yaml

# Monitor agent
tli agent status
tli agent performance --window 24h
tli agent activity stream
```

### 4. Backtesting Service ↔ Trading Agent

**Backtesting simulates Trading Agent**:

```rust
// Backtesting Service simulates Trading Agent decisions
struct BacktestingAgent {
    agent_service_client: TradingAgentServiceClient,
    simulated_time: DateTime<Utc>,
}

impl BacktestingAgent {
    async fn run_backtest(&self, historical_data: Vec<BarData>) -> BacktestResult {
        for bar in historical_data {
            // Simulate universe selection
            let universe = self.agent_service_client
                .select_universe(SelectUniverseRequest {
                    criteria: default_criteria(),
                    ..Default::default()
                })
                .await?;

            // Simulate asset selection
            let assets = self.agent_service_client
                .select_assets(SelectAssetsRequest {
                    universe_id: universe.universe_id,
                    ..Default::default()
                })
                .await?;

            // Simulate allocation
            let allocation = self.agent_service_client
                .allocate_portfolio(AllocatePortfolioRequest {
                    assets: assets.assets,
                    ..Default::default()
                })
                .await?;

            // Simulate order generation
            let orders = self.agent_service_client
                .generate_orders(GenerateOrdersRequest {
                    allocation_id: allocation.allocation_id,
                    ..Default::default()
                })
                .await?;

            // Track simulated results
            self.apply_orders_to_simulation(orders);
        }

        Ok(self.calculate_backtest_metrics())
    }
}
```

---

## Service Configuration

### Port Allocation

| Service | gRPC Port | Health Port | Metrics Port |
|---------|-----------|-------------|--------------|
| API Gateway | 50051 | 8080 | 9091 |
| Trading Service | 50052 | 8081 | 9092 |
| Backtesting Service | 50053 | 8082 | 9093 |
| ML Training Service | 50054 | 8095 | 9094 |
| **Trading Agent Service** | **50055** | **8083** | **9095** |

### Environment Variables

```bash
# Trading Agent Service Configuration
TRADING_AGENT_SERVICE_HOST=0.0.0.0
TRADING_AGENT_SERVICE_PORT=50055
TRADING_AGENT_HEALTH_PORT=8083
TRADING_AGENT_METRICS_PORT=9095

# Integration Configuration
TRADING_SERVICE_URL=http://trading_service:50052
ML_TRAINING_SERVICE_URL=http://ml_training_service:50054

# Agent Configuration
AGENT_UNIVERSE_REFRESH_INTERVAL=3600  # seconds (1 hour)
AGENT_ASSET_SELECTION_INTERVAL=300    # seconds (5 minutes)
AGENT_REBALANCE_THRESHOLD=0.05        # 5% drift triggers rebalance
AGENT_DEFAULT_CAPITAL=1000000         # $1M default capital

# Risk Configuration
AGENT_MAX_POSITION_SIZE_PCT=0.20      # 20% max per position
AGENT_MAX_LEVERAGE=2.0                # 2x max leverage
AGENT_MAX_PORTFOLIO_VAR_95=0.05       # 5% max VaR (95%)

# Database
DATABASE_URL=postgresql://foxhunt:foxhunt_dev_password@postgres:5432/foxhunt

# Monitoring
RUST_LOG=info
PROMETHEUS_ENABLED=true
```

### Docker Compose Entry

```yaml
trading_agent_service:
  build:
    context: .
    dockerfile: services/trading_agent_service/Dockerfile
  ports:
    - "50055:50055"  # gRPC
    - "8083:8083"    # Health
    - "9095:9095"    # Metrics
  environment:
    - TRADING_AGENT_SERVICE_PORT=50055
    - TRADING_SERVICE_URL=http://trading_service:50052
    - ML_TRAINING_SERVICE_URL=http://ml_training_service:50054
    - DATABASE_URL=postgresql://foxhunt:foxhunt_dev_password@postgres:5432/foxhunt
    - RUST_LOG=info
  depends_on:
    - postgres
    - trading_service
    - ml_training_service
  networks:
    - foxhunt_network
  healthcheck:
    test: ["CMD", "grpc_health_probe", "-addr=:50055"]
    interval: 10s
    timeout: 5s
    retries: 3
```

---

## Database Schema

### Tables

```sql
-- Universe history
CREATE TABLE trading_universes (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    universe_id TEXT NOT NULL UNIQUE,
    criteria JSONB NOT NULL,
    instruments JSONB NOT NULL,  -- Array of Instrument objects
    metrics JSONB NOT NULL,      -- UniverseMetrics
    created_at TIMESTAMPTZ DEFAULT NOW(),
    updated_at TIMESTAMPTZ DEFAULT NOW()
);

CREATE INDEX idx_universes_created_at ON trading_universes(created_at DESC);

-- Asset selection history
CREATE TABLE asset_selections (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    universe_id TEXT REFERENCES trading_universes(universe_id),
    criteria JSONB NOT NULL,
    asset_scores JSONB NOT NULL,  -- Array of AssetScore objects
    metrics JSONB NOT NULL,
    selected_at TIMESTAMPTZ DEFAULT NOW()
);

CREATE INDEX idx_asset_selections_universe ON asset_selections(universe_id);
CREATE INDEX idx_asset_selections_selected_at ON asset_selections(selected_at DESC);

-- Portfolio allocations
CREATE TABLE portfolio_allocations (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    allocation_id TEXT NOT NULL UNIQUE,
    strategy JSONB NOT NULL,       -- AllocationStrategy
    risk_constraints JSONB NOT NULL,
    allocations JSONB NOT NULL,    -- Array of AssetAllocation objects
    metrics JSONB NOT NULL,
    total_capital NUMERIC(20, 2),
    created_at TIMESTAMPTZ DEFAULT NOW()
);

CREATE INDEX idx_allocations_created_at ON portfolio_allocations(created_at DESC);

-- Order batches
CREATE TABLE order_batches (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    order_batch_id TEXT NOT NULL UNIQUE,
    allocation_id TEXT REFERENCES portfolio_allocations(allocation_id),
    orders JSONB NOT NULL,         -- Array of GeneratedOrder objects
    metrics JSONB NOT NULL,
    submission_results JSONB,      -- Array of OrderSubmissionResult objects
    created_at TIMESTAMPTZ DEFAULT NOW(),
    submitted_at TIMESTAMPTZ
);

CREATE INDEX idx_order_batches_created_at ON order_batches(created_at DESC);
CREATE INDEX idx_order_batches_allocation ON order_batches(allocation_id);

-- Registered strategies
CREATE TABLE agent_strategies (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    strategy_id TEXT NOT NULL UNIQUE,
    strategy_name TEXT NOT NULL,
    strategy_type TEXT NOT NULL,
    status TEXT NOT NULL,          -- ENABLED, DISABLED, PAUSED, ERROR
    config JSONB NOT NULL,
    performance JSONB,             -- StrategyPerformance
    created_at TIMESTAMPTZ DEFAULT NOW(),
    updated_at TIMESTAMPTZ DEFAULT NOW()
);

CREATE INDEX idx_strategies_status ON agent_strategies(status);
CREATE INDEX idx_strategies_name ON agent_strategies(strategy_name);

-- Agent activity log
CREATE TABLE agent_activity_log (
    id BIGSERIAL PRIMARY KEY,
    activity_type TEXT NOT NULL,   -- UNIVERSE_SELECTION, ASSET_SELECTION, etc.
    event_data JSONB NOT NULL,
    timestamp TIMESTAMPTZ DEFAULT NOW()
);

CREATE INDEX idx_activity_log_timestamp ON agent_activity_log(timestamp DESC);
CREATE INDEX idx_activity_log_type ON agent_activity_log(activity_type);

-- Agent performance metrics (time-series)
CREATE TABLE agent_performance_metrics (
    id BIGSERIAL PRIMARY KEY,
    metrics JSONB NOT NULL,        -- AgentPerformanceMetrics
    period_start TIMESTAMPTZ NOT NULL,
    period_end TIMESTAMPTZ NOT NULL,
    recorded_at TIMESTAMPTZ DEFAULT NOW()
);

CREATE INDEX idx_performance_metrics_period ON agent_performance_metrics(period_start, period_end);
```

---

## Implementation Plan

### Phase 1: Core Service Setup (Week 1)

1. **Create Service Skeleton**:
   - Initialize `services/trading_agent_service/` directory structure
   - Setup `Cargo.toml` with dependencies
   - Create `proto/trading_agent.proto`
   - Generate gRPC stubs

2. **Implement Basic gRPC Server**:
   - `main.rs` with gRPC server setup
   - Health check endpoint
   - Prometheus metrics integration

3. **Database Setup**:
   - Create migration for Trading Agent tables
   - Implement repository traits
   - Setup connection pooling

4. **Docker Integration**:
   - Create `Dockerfile`
   - Add to `docker-compose.yml`
   - Configure networking

### Phase 2: Universe & Asset Selection (Week 2)

1. **Universe Selection**:
   - Implement `SelectUniverse` logic
   - Market data integration
   - Volatility calculation
   - Liquidity scoring
   - Store universe in database

2. **Asset Selection**:
   - Implement `SelectAssets` logic
   - ML signal integration (call ML Training Service)
   - Factor score calculation
   - Composite scoring algorithm

3. **Testing**:
   - Unit tests for selection logic
   - Integration tests with mock ML service

### Phase 3: Portfolio Allocation (Week 3)

1. **Allocation Strategies**:
   - Equal-weight allocation
   - Risk-parity allocation
   - ML-optimized allocation
   - Kelly criterion

2. **Risk Constraint Engine**:
   - Position size limits
   - Sector exposure limits
   - VaR calculation
   - Leverage checks

3. **Rebalancing Logic**:
   - Drift detection
   - Rebalance action generation
   - Cost estimation

### Phase 4: Order Generation & Execution (Week 4)

1. **Order Generation**:
   - Delta calculation (target - current)
   - Order type selection (market/limit)
   - Price determination
   - Order metadata tagging

2. **Trading Service Integration**:
   - gRPC client for Trading Service
   - Order submission logic
   - Result handling and logging

3. **Testing**:
   - End-to-end tests with Trading Service
   - Paper trading simulation

### Phase 5: Strategy Coordination (Week 5)

1. **Strategy Framework**:
   - Strategy registration
   - Strategy lifecycle management
   - Performance tracking per strategy

2. **Built-in Strategies**:
   - ML Ensemble strategy
   - Mean reversion strategy
   - Momentum strategy

3. **Strategy Execution Engine**:
   - Periodic execution scheduler
   - Strategy isolation
   - Error handling

### Phase 6: Monitoring & API Gateway Integration (Week 6)

1. **Agent Monitoring**:
   - Real-time status API
   - Activity streaming
   - Performance metrics calculation

2. **API Gateway Integration**:
   - Add Trading Agent proxy to API Gateway
   - Update TLI with agent commands
   - Documentation

3. **Observability**:
   - Grafana dashboard for agent metrics
   - Prometheus alerts for agent errors
   - Structured logging

### Phase 7: Backtesting Integration (Week 7)

1. **Backtesting Simulation**:
   - Backtest adapter for Trading Agent
   - Historical replay logic
   - Performance comparison

2. **Testing & Validation**:
   - End-to-end tests across all services
   - Load testing
   - Chaos testing (service failures)

### Phase 8: Production Hardening (Week 8)

1. **Error Handling**:
   - Graceful degradation
   - Circuit breakers
   - Retry logic

2. **Performance Optimization**:
   - Database query optimization
   - Caching strategies
   - Connection pooling tuning

3. **Documentation**:
   - API documentation
   - Deployment guide
   - Operational runbook

---

## Success Criteria

### Functional Requirements

- ✅ Universe selection completes in <1 second
- ✅ Asset selection completes in <2 seconds (including ML query)
- ✅ Portfolio allocation completes in <500ms
- ✅ Order generation completes in <200ms
- ✅ End-to-end (universe → orders) completes in <5 seconds
- ✅ Strategies execute on schedule with <100ms jitter
- ✅ All APIs return in <100ms (excluding long-running operations)

### Non-Functional Requirements

- ✅ Service uptime >99.9%
- ✅ No data loss (all decisions logged to database)
- ✅ Prometheus metrics exported
- ✅ Health checks respond in <10ms
- ✅ Graceful shutdown (drain in-flight requests)
- ✅ Docker container restart recovery

### Integration Requirements

- ✅ Trading Service integration (order submission)
- ✅ ML Training Service integration (prediction queries)
- ✅ API Gateway proxy configured
- ✅ TLI commands functional
- ✅ Backtesting simulation working

### Testing Requirements

- ✅ Unit test coverage >80%
- ✅ Integration tests for all gRPC methods
- ✅ End-to-end tests across services
- ✅ Load tests (100 req/s sustained)
- ✅ Chaos tests (service failure recovery)

---

## Risk Analysis

### Technical Risks

| Risk | Impact | Probability | Mitigation |
|------|--------|-------------|------------|
| ML service latency | High | Medium | Cache predictions, use stale data if needed |
| Trading service downtime | Critical | Low | Queue orders, retry with exponential backoff |
| Database bottleneck | High | Medium | Index optimization, read replicas, caching |
| Strategy logic errors | Critical | Medium | Extensive testing, paper trading validation |
| Order submission failures | High | Medium | Idempotent retry, comprehensive error handling |

### Operational Risks

| Risk | Impact | Probability | Mitigation |
|------|--------|-------------|------------|
| Configuration errors | High | Medium | Schema validation, default values, dry-run mode |
| Resource exhaustion | High | Low | Resource limits, monitoring alerts |
| Data corruption | Critical | Low | Database transactions, audit logging |
| Version incompatibility | Medium | Low | API versioning, backward compatibility |

---

## Alternatives Considered

### Alternative 1: Embed Agent Logic in Trading Service
**Pros**: Simpler architecture, lower latency
**Cons**: Violates SRP, harder to test, couples decision-making with execution
**Decision**: ❌ Rejected - doesn't scale, poor separation of concerns

### Alternative 2: Use Message Queue Instead of gRPC
**Pros**: Decoupling, buffering, retry semantics
**Cons**: Added complexity, harder to debug, eventual consistency
**Decision**: ❌ Rejected for MVP - can add later if needed

### Alternative 3: Agent as Library, Not Service
**Pros**: No network overhead, simpler deployment
**Cons**: Can't reuse across services, harder to version independently
**Decision**: ❌ Rejected - limits reusability (backtesting needs it)

---

## Future Enhancements

### Phase 2 (Post-MVP)

1. **Advanced Allocation Strategies**:
   - Black-Litterman allocation
   - Hierarchical risk parity
   - Reinforcement learning-based allocation

2. **Multi-Account Support**:
   - Manage multiple trading accounts
   - Cross-account risk aggregation

3. **Regime Detection**:
   - Automatic strategy switching based on market regime
   - Volatility regime detection

4. **Advanced Rebalancing**:
   - Tax-aware rebalancing
   - Transaction cost optimization

5. **Strategy Marketplace**:
   - User-defined strategies
   - Strategy backtesting UI
   - Strategy performance leaderboard

---

## Appendix A: Example Workflow

**Scenario**: Daily portfolio rebalancing at market open

```
1. 08:30 AM: Universe selection job triggers
   → SelectUniverse(criteria: {min_liquidity: 0.7, max_volatility: 0.5})
   → Returns: [ES.FUT, NQ.FUT, ZN.FUT, 6E.FUT]

2. 08:31 AM: Asset selection job triggers
   → SelectAssets(universe_id: "univ-123", max_assets: 3)
   → Queries ML Training Service for predictions
   → Returns: [ES.FUT (score: 0.85), NQ.FUT (score: 0.78), ZN.FUT (score: 0.72)]

3. 08:32 AM: Portfolio allocation job triggers
   → AllocatePortfolio(assets: [...], strategy: RISK_PARITY, capital: $1M)
   → Returns: {ES.FUT: 35%, NQ.FUT: 40%, ZN.FUT: 25%}

4. 08:33 AM: Order generation job triggers
   → GenerateOrders(allocation_id: "alloc-456", ml_signals: [...])
   → Returns: [BUY ES.FUT 15 @ MARKET, BUY NQ.FUT 20 @ MARKET, SELL ZN.FUT 5 @ MARKET]

5. 08:33 AM: Order submission
   → SubmitAgentOrders(order_batch_id: "batch-789", orders: [...])
   → Calls Trading Service.SubmitMLOrder() for each order
   → Returns: {accepted: 3, rejected: 0, acceptance_rate: 1.0}

6. 08:34 AM: Monitor execution
   → StreamAgentActivity() streams order fill events
   → Logs execution results to database
   → Updates performance metrics

7. 08:35 AM: Performance tracking
   → GetAgentPerformance(window: 24h)
   → Returns: {total_pnl: $12,500, sharpe_ratio: 1.8, win_rate: 0.65}
```

---

## Appendix B: Key Metrics

### Trading Agent Metrics (Prometheus)

```prometheus
# Universe selection
trading_agent_universe_size{universe_id} gauge
trading_agent_universe_refresh_duration_seconds histogram
trading_agent_universe_liquidity_score{universe_id} gauge

# Asset selection
trading_agent_selected_assets{universe_id} gauge
trading_agent_asset_selection_duration_seconds histogram
trading_agent_asset_composite_score{symbol} gauge

# Portfolio allocation
trading_agent_portfolio_utilization gauge  # % capital deployed
trading_agent_portfolio_volatility gauge
trading_agent_portfolio_sharpe gauge
trading_agent_allocation_duration_seconds histogram

# Order generation
trading_agent_orders_generated counter
trading_agent_orders_submitted counter
trading_agent_orders_accepted counter
trading_agent_orders_rejected counter
trading_agent_order_acceptance_rate gauge

# Strategy performance
trading_agent_strategy_pnl{strategy_id} gauge
trading_agent_strategy_sharpe{strategy_id} gauge
trading_agent_strategy_trades{strategy_id} counter

# Agent health
trading_agent_active_strategies gauge
trading_agent_errors_total{error_type} counter
trading_agent_api_request_duration_seconds{method} histogram
```

---

## Conclusion

The **Trading Agent Service** is a critical component that separates trading decision-making from execution. By following this design, we achieve:

1. **Separation of Concerns**: Decision-making (Agent) vs. execution (Trading Service)
2. **Reusability**: Backtesting can reuse agent logic
3. **Testability**: Agent logic can be tested independently
4. **Scalability**: Agent can be scaled independently of Trading Service
5. **Maintainability**: Clear boundaries between components

**Next Steps**:
1. Review and approve this design
2. Create GitHub issues for each implementation phase
3. Start with Phase 1: Core Service Setup
4. Iterative development with weekly demos

---

**Document Status**: ✅ **READY FOR REVIEW**
**Estimated Implementation Time**: 8 weeks (1 developer)
**Dependencies**: Trading Service, ML Training Service, API Gateway
**Risk Level**: Medium (new service, but clear interfaces)