Skip to main content
All topics

AI Orchestrator

How the Multi-Agent Swarm and Strategy Factory manage your portfolio autonomously.

Multi-Agent Swarm

The Multi-Agent Swarm is Smartbull's core decision-making system. It consists of 5 specialized AI agents:

  • Trend Agent — detects and follows momentum across timeframes
  • Mean-Reversion Agent — identifies overextended moves for counter-trend entries
  • Volatility Agent — trades volatility expansion/compression cycles
  • Macro Agent — incorporates DXY, yields, VIX, and funding rates
  • Sentiment Agent — processes news, social media, and on-chain signals

Each agent independently analyzes the market and produces a signal with confidence score. A consensus mechanism aggregates votes — requiring majority agreement before any trade is executed. This eliminates single-point-of-failure risk.

Strategy Factory

The Strategy Factory is an automated strategy generation pipeline that runs 24/7:

  • Hypothesis Generation — LLM creates new trading strategy ideas based on market patterns
  • Compilation — strategies are compiled into executable code in a sandboxed environment
  • Backtesting — each strategy is walk-forward tested on 2+ years of data
  • 20% Filter — only strategies with annualized return ≥20% pass to the next stage
  • Decorrelation Check — rejects strategies correlated >0.6 with existing ones
  • Paper Trading — 14-day live paper test before any real capital allocation
  • Promotion — successful strategies are promoted to shadow, then live

This creates a continuous pipeline of fresh alpha sources without manual intervention.

Regime-Aware Meta-Learner v2

The Meta-Learner v2 uses a Hidden Markov Model (HMM) to detect 4 market regimes:

  • Trending — strong directional moves, momentum strategies get more weight
  • Mean-Reverting — range-bound, counter-trend strategies dominate
  • Volatile — high uncertainty, position sizes reduced, hedges activated
  • Quiet — low vol, carry/yield strategies preferred

Weights are dynamically adjusted every tick based on the detected regime. Transitions are smoothed to avoid whipsawing.

Strategy Lifecycle

Every strategy goes through a strict lifecycle:

  • Candidate — newly generated, awaiting backtest validation
  • Paper — 14-day paper trading experiment
  • Shadow — live signals tracked but no real orders
  • Live — actively trading with real capital
  • Deprecated — underperforming, being phased out
  • Retired — removed from active rotation

You can monitor the full lifecycle on your dashboard. The system automatically promotes and retires strategies based on performance.