Skip to main content

AI Agents Overview

Status: Verified against code Last Updated: 2025-12-11

What This Section Covers

The TTG algorithm uses a multi-agent AI system where specialized agents analyze different aspects of a trade, then a supervisor synthesizes their opinions into a final decision.

System Architecture

The system consists of 5 agents working together:

AgentTypeRoleWhat It Does
MIKE AgentOrchestratorCoordinatorOrchestrates the entire workflow, collects data, dispatches to specialists
Technical AgentSpecialistChart AnalystPrice action, patterns, S/R levels, indicators
Macro AgentSpecialistBig PictureMarket regime, cross-asset, economic data, news
Wild Card AgentSpecialistRisk WatchdogExternal events, timing risks, contingencies
Supervisor AgentSpecialistFinal DecisionSynthesizes all verdicts, picks options contracts

MIKE Agent (Orchestrator)

MIKE is the Multi-Agent Trading Intelligence Orchestrator. It doesn't analyze markets — it coordinates the other agents.

What MIKE does:

  1. Receives user request (ticker, asset type, trade style, account size, risk %)
  2. Calls Data Collection Service to gather market data
  3. Dispatches Technical, Macro, and Wild Card agents in parallel
  4. Passes all verdicts to Supervisor for final synthesis
  5. Returns complete trade plan

How They Work Together

User Request (ticker, style, account size, risk %)


┌─────────────────────┐
│ MIKE Agent │
│ (Orchestrator) │
└─────────────────────┘


Data Collection Service

┌───────────┼───────────┐
▼ ▼ ▼
Technical Macro Wild Card
Agent Agent Agent
(parallel) (parallel) (parallel)
│ │ │
└───────────┴───────────┘


┌─────────────────────┐
│ Supervisor Agent │
│ (Final Decision) │
└─────────────────────┘


Trade Plan Output

Key Principles

1. Parallel Execution

The Technical, Macro, and Wild Card agents run simultaneously (not sequentially). This speeds up analysis.

2. Independence

Each agent forms its own opinion without knowing what the others think. This prevents groupthink.

  • Technical Agent only sees price/indicator data
  • Macro Agent only sees market context/news
  • Wild Card Agent only sees risk scenarios

3. Conflict Resolution

When agents disagree, Technical Agent wins because it has actual price data.

Rule: When agents conflict, go with the Technical Agent (it has actual price action).

4. Default to Trade

The system is biased toward trading:

"YOU ARE A TRADER. TRADERS TRADE. Default action is TO TRADE."

Trade or No-Trade Rules

ALWAYS TRADE when:

  • Both Technical and Macro agree (MANDATORY)
  • ONE agent is directional with >50% confidence
  • BOTH agents have ANY directional opinion (trade the stronger one)

NO TRADE only when:

  • BOTH agents say "neutral" with <40% confidence EACH
  • Market is completely closed (weekend only)
  • Critical system error or all data missing

Data Flow

┌──────────────────┐
│ User Request │
│ ticker, style, │
│ account, risk % │
└────────┬─────────┘


┌──────────────────┐
│ Data Collection │
│ Service │──────────────────────────────┐
│ │ │
│ - Snapshots │ │
│ - Historical bars│ │
│ - Indicators │ │
│ - Options chain │ │
│ - News │ │
│ - Economic data │ │
│ - International │ │
└────────┬─────────┘ │
│ │
│ market_data │
│ │
▼ │
┌──────────────────┐ │
│ Technical Agent │───┐ │
│ direction, │ │ │
│ confidence, │ │ │
│ S/R levels │ │ │
└──────────────────┘ │ │
│ │
┌──────────────────┐ │ │
│ Macro Agent │───┼─► Supervisor Agent ◄─────┘
│ market_regime, │ │ (receives full │
│ news_sentiment │ │ market_data for │
└──────────────────┘ │ options selection) │
│ │
┌──────────────────┐ │ │
│ Wild Card Agent │───┘ │
│ risks, │ │
│ contingencies │ │
└──────────────────┘ │

▼ │
┌──────────────────┐ │
│ Trade Plan │ │
│ - Direction │ │
│ - Entry zone │ │
│ - Position size │ │
│ - Stop/targets │ │
│ - Contract (opts)│ │
│ - Wild cards │ │
└──────────────────┘

LLM Configuration

All agents use the same underlying language model:

  • Model: Google Gemini 3 Pro (via secure API gateway)
  • Configuration: Optimized for trading analysis with high reasoning capacity

Provider routing and specific model parameters are intentionally not documented here.

Performance Metrics (Transparency)

For complete transparency about how we measure performance:

Win Rate Calculation

Win rate is calculated as:

Win Rate = WINS ÷ (WINS + LOSSES) × 100

This measures directional accuracy — did the trade hit its target before hitting its stop?

Outcome Categories

Not all trade plans result in a clear WIN or LOSS. Here's what can happen:

OutcomeIncluded in Win Rate?What It Means
WIN✅ Yes (numerator)Target was hit before stop
LOSS✅ Yes (denominator)Stop was hit before target
NO_ENTRY❌ NoPrice never reached entry zone
EXPIRED❌ NoOptions contract expired (for time-based evaluation)
PENDING❌ NoTrade still open, not yet evaluated
ERROR❌ NoSystem error during evaluation
SKIPPED❌ NoTrade plan was NO TRADE recommendation

Why Some Outcomes Are Excluded

  • NO_ENTRY trades never got filled — the system correctly identified a zone that wasn't reached
  • EXPIRED outcomes apply to evaluation windows, not actual trade execution
  • PENDING trades haven't concluded yet
  • SKIPPED means the system recommended not trading (this is actually a feature, not a failure)

What This Means

The published win rate reflects decisive outcomes only (trades that clearly won or lost). This is the most honest representation of directional accuracy.

If you want the "full funnel" view:

  • Total trade plans generated
  • Minus NO TRADE recommendations
  • Minus NO_ENTRY (never filled)
  • Equals trades that actually executed
  • Of those, WIN ÷ (WIN + LOSS) = Win Rate

Options Trades Note

For OPTIONS trades specifically, win/loss is evaluated based on underlying stock movement hitting the target or stop price. See the Options Data page for details on this methodology.

Core Components

Agents:

  • MIKE Agent (Orchestrator) — Coordinates the workflow and dispatches to specialists
  • Technical Agent — Price action, patterns, S/R levels, indicators (when available)
  • Macro Agent — Market regime, cross-asset context, economic/news impact
  • Wild Card Agent — External event/timing risks, contingencies
  • Supervisor Agent — Synthesizes verdicts into the final plan (and selects options contracts)

Supporting services:

  • Market data collection — Gathers snapshot/bars/chain/news inputs
  • Trade style configuration — Controls lookbacks, bars-to-AI, and indicator windows