Strategy Variants

Overview

Autonome uses a multi-variant strategy architecture that allows multiple AI trading personalities to operate in parallel. Each variant represents a distinct trading philosophy, risk tolerance, and decision-making approach optimized for different market regimes.

The Four Variants

Autonome currently supports four strategy variants, each designed for specific market conditions:

Apex (Kelly Engine)

Aggressive 10x leverage strategy using volatility squeezes and VWAP momentum validation

Trendsurfer (Momentum)

Trend-following strategy that rides strong directional moves using ADX filtering and Kijun-Sen trailing stops

Contrarian (Reverter)

Mean reversion specialist that fades extremes in ranging markets when ADX < 25

Sovereign (Adaptive)

Balanced regime-adaptive allocator that blends trend and range strategies based on market conditions

Variant Architecture

Code Organization

The variant system is built on a centralized configuration model:

// Core variant definitions
src/core/shared/variants/index.ts        // SSOT for variant IDs, configs, styling

// Prompt files (each exports SYSTEM_PROMPT and USER_PROMPT)
src/server/features/trading/prompts/
  ├── apex.ts                             // Apex strategy prompts
  ├── trendsurfer.ts                      // Trendsurfer prompts
  ├── contrarian.ts                       // Contrarian prompts
  ├── sovereign.ts                        // Sovereign prompts
  ├── variants.ts                         // Maps variants to prompts
  ├── promptBuilder.ts                    // Assembles prompts with data
  └── promptSections.ts                   // Builds portfolio/performance sections

Single Source of Truth

All variant metadata lives in src/core/shared/variants/index.ts:

export const VARIANT_IDS = [
  "Apex",
  "Trendsurfer", 
  "Contrarian",
  "Sovereign",
] as const;

export const VARIANT_CONFIG: Record<VariantId, VariantConfig> = {
  Apex: {
    id: "Apex",
    label: "Apex (Kelly Engine)",
    description: "Aggressive 10x leverage, VWAP momentum validation, squeeze trading",
    color: "#a855f7",
    bgClass: "bg-purple-500/20",
    textClass: "text-purple-600",
    lightBg: "#faf5ff",
  },
  // ... other variants
};

This configuration is consumed by:

Database schema (variant enum)
oRPC router schemas (Zod validation)
Frontend components (styling, filtering)
Seed scripts
Export utilities

Prompt System

Each variant has two core components: 1. SYSTEM_PROMPT (Static Instructions)

Trading philosophy and identity
Decision framework
Tool interface documentation
Risk management rules
Exit plan requirements
Response format guidelines

2. USER_PROMPT (Dynamic Data Template)

Session metadata (time, invocations, cash)
Market intelligence
Portfolio snapshot
Open positions table
Performance overview
Mission/instructions

Prompts are built using promptBuilder.ts:buildTradingPrompts():

const { systemPrompt, userPrompt } = buildTradingPrompts({
  account,
  portfolio,
  openPositions,
  exposureSummary,
  performanceMetrics,
  marketIntelligence,
  currentTime,
  variant: "Apex", // or "Trendsurfer", "Contrarian", "Sovereign"
});

Data Provided to AI

All variants receive identical market data and portfolio state, but interpret it differently based on their strategy:

Session Header

Session: 120 min | Interval: 5 min | Invocations: 24 | 2:30 PM IST
Cash: $8,342.15 | Exposure: 45.2%

Market Intelligence

Provided via {{MARKET_INTELLIGENCE}} placeholder:

Technical indicators (Bollinger Bands, VWAP, EMA20, ADX, Supertrend, Ichimoku)
Volume data (current vs average)
Funding rates
Price action context

Portfolio Snapshot

Built by promptSections.ts:buildPortfolioSnapshotSection():

portfolio_value: $10,234.56 | available_cash: $5,621.33 | open_positions: 2
cash_utilization_pct: 45.1% | exposure_to_equity_pct: 45.1%
gross_exposure_usd: $4,613.23 | long_exposure: $4,613.23 | short_exposure: $0.00 | net_exposure: $4,613.23
unrealized_pnl: $234.56 | scaled_realized_pnl: $0.00
gross_risk_usd: $128.45 | risk_to_equity_pct: 1.25% | max_single_position_risk_usd: $85.32 | max_single_position_risk_pct: 0.83%

Open Positions

Built by promptSections.ts:buildOpenPositionsSection():

symbol BTCUSDT | side LONG | qty 0.2500 | entry 42150.00 | mark 42380.00 | notional $2,119.00 | leverage 5.00x
unrealized $57.50 | scaled_realized $0.00
risk_usd $85.32 | risk_pct 0.83% | reward_usd $211.50 | reward_pct 2.06% | rr_ratio 2.48
exit_plan: target 42800.00 | stop 41809.00 | invalidation "Close below EMA20" | time_exit "Close if held > 12h" | cooldown_until 2026-03-07T14:35:00Z
intent: signal LONG | confidence 75.0% | decision_status active | last_decision_at 2026-03-07T14:30:00Z

Performance Metrics

Built by promptSections.ts:buildPerformanceOverview():

closed_trade_realized_pnl: $423.45 | trade_count: 12 | win_rate: 58.33%
total_return_since_start: 4.23% | annualized_sharpe_ratio: 1.85
current_drawdown: -2.15% | max_drawdown: -5.42%

Prompt Data Principles

The system follows strict guidelines for prompt construction:

Spoon-feed data: AI should never infer or calculate - provide all metrics explicitly
Explicit labels: Use risk_usd $128.56 not risk $128.56 - no ambiguity
Show zeros: scaled_realized $0.00 is meaningful (no partial closes yet)
Omit N/A: If data doesn’t exist, don’t show it (noise reduction)
No token optimization: Clarity > brevity. Full descriptive labels always
Section separation: Header (session), PORTFOLIO (current state), PERFORMANCE (historical), OPEN POSITIONS (per-position)
No duplication: Each metric lives in exactly one section

Source: promptSections.ts:1-176

Tool Interface

All variants use the same tool interface for portfolio control:

createPosition: Open new positions with leverage, stops, targets
closePosition: Exit positions
updateExitPlan: Modify stops/targets on open positions
holding: Explicit no-action with reasoning

Critical Rule: Every response must end with a tool call. If no action is needed, call holding() with reasoning.

Common Constraints

All variants share these core constraints:

Data Source Hierarchy

Manual/Exchange Indicators (Execution): Use for exact entry price, stop loss, and invalidation
Taapi/Binance Indicators (Context): Use for broad trend and market regime (ADX, Supertrend, Ichimoku)

Mandatory Exit Plan

Every position must specify:

invalidation_condition: The thesis-killing event (e.g., “Close below VWAP”)
invalidation_price: The exact stop-loss price
time_exit: Max hold duration (e.g., “Close if held > 12h without profit”)
cooldown_minutes: 1-15 minutes to prevent impulsive flips

Cooldown System

Cooldown prevents direction flips. After closing a LONG position, the AI cannot open a SHORT position on the same symbol until cooldown expires (and vice versa). This applies both while holding AND after closing.

Variant Comparison

Risk Tolerance
Entry Triggers
Exit Strategy

Variant	Max Leverage	Risk Profile	Best For
Apex	10x	Extremely Aggressive	High-conviction squeezes
Trendsurfer	5x	Moderate	Strong trends (ADX > 25)
Contrarian	5x	Moderate	Ranging markets (ADX < 25)
Sovereign	5x	Balanced	All market conditions

Variant	Primary Setup	Filter	Confirmation
Apex	Bollinger Squeeze	Price vs VWAP	Breakout direction
Trendsurfer	20-period High breakout	ADX > 25, Cloud position	Trend strength
Contrarian	Bollinger Band touch	ADX < 25, Low volume	Rejection wick
Sovereign	Regime-adaptive	ADX + Cloud	Context-dependent

Variant	Stop Management	Profit Taking	Time Exit
Apex	Profit Ratchet (move to BE at +15% ROE)	Trail at +30%	2 hours
Trendsurfer	Trail using Kijun-Sen or EMA20	Never use fixed targets	No fixed time
Contrarian	1x ATR beyond BB	Target VWAP	12 hours
Sovereign	Thesis-based invalidation	Context-dependent	12 hours

Adding a New Variant

To add a new trading strategy variant:

Add to SSOT: Update src/core/shared/variants/index.ts

export const VARIANT_IDS = [
  "Apex",
  "Trendsurfer",
  "Contrarian",
  "Sovereign",
  "YourNewVariant", // Add here
] as const;

export const VARIANT_CONFIG: Record<VariantId, VariantConfig> = {
  // ... existing variants
  YourNewVariant: {
    id: "YourNewVariant",
    label: "Your New Variant (Description)",
    description: "Strategy description",
    color: "#hexcolor",
    bgClass: "bg-color-500/20",
    textClass: "text-color-600",
    lightBg: "#hexcolor",
  },
};

Create Prompt File: Add src/server/features/trading/prompts/yournewvariant.ts

export const SYSTEM_PROMPT = `You are **Autonome YourNewVariant**...
// Strategy instructions
`;

export const USER_PROMPT = `
Session: {{TOTAL_MINUTES}} min | Interval: 5 min | Invocations: {{INVOKATION_TIMES}}
// Template with placeholders
`;

Register in Variants: Update src/server/features/trading/prompts/variants.ts

import {
  SYSTEM_PROMPT as SYSTEM_PROMPT_YOURNEWVARIANT,
  USER_PROMPT as USER_PROMPT_YOURNEWVARIANT,
} from "./yournewvariant";

export const VARIANT_PROMPTS: Record<VariantId, VariantConfig> = {
  // ... existing variants
  YourNewVariant: {
    ...VARIANT_CONFIG.YourNewVariant,
    systemPrompt: SYSTEM_PROMPT_YOURNEWVARIANT,
    userPrompt: USER_PROMPT_YOURNEWVARIANT,
    temperature: 0,
  },
};

Migrate Database: Run bun run db:generate && bun run db:migrate

Next Steps

Apex Strategy

Learn about the aggressive 10x leverage Kelly Engine

Trendsurfer Strategy

Explore the momentum-based trend follower

Contrarian Strategy

Understand the mean reversion specialist

Sovereign Strategy

Discover the balanced adaptive allocator

Trading System

AI Strategies

Simulator

Strategy Variants

Overview

The Four Variants

Apex (Kelly Engine)

Trendsurfer (Momentum)

Contrarian (Reverter)

Sovereign (Adaptive)

Variant Architecture

Code Organization

Single Source of Truth

Prompt System

Data Provided to AI

Session Header

Market Intelligence

Portfolio Snapshot

Open Positions

Performance Metrics

Prompt Data Principles

Tool Interface

Common Constraints

Data Source Hierarchy

Mandatory Exit Plan

Cooldown System

Variant Comparison

Adding a New Variant

Next Steps

Apex Strategy

Trendsurfer Strategy

Contrarian Strategy

Sovereign Strategy

Build docs developers (and LLMs) love

Trading System

AI Strategies

Simulator

​Overview

​The Four Variants

Apex (Kelly Engine)

Trendsurfer (Momentum)

Contrarian (Reverter)

Sovereign (Adaptive)

​Variant Architecture

​Code Organization

​Single Source of Truth

​Prompt System

​Data Provided to AI

​Session Header

​Market Intelligence

​Portfolio Snapshot

​Open Positions

​Performance Metrics

​Prompt Data Principles

​Tool Interface

​Common Constraints

​Data Source Hierarchy

​Mandatory Exit Plan

​Cooldown System

​Variant Comparison

​Adding a New Variant

​Next Steps

Apex Strategy

Trendsurfer Strategy

Contrarian Strategy

Sovereign Strategy

Build docs developers (and LLMs) love

Overview

The Four Variants

Variant Architecture

Code Organization

Single Source of Truth

Prompt System

Data Provided to AI

Session Header

Market Intelligence

Portfolio Snapshot

Open Positions

Performance Metrics

Prompt Data Principles

Tool Interface

Common Constraints

Data Source Hierarchy

Mandatory Exit Plan

Cooldown System

Variant Comparison

Adding a New Variant

Next Steps