Build trading strategies with code, not clicks.

Introduction

Most trading platforms make you click through a form: pick an indicator, set thresholds, save. ThunderScript flips that. You declare your strategy as data: indicators as variables, conditions as boolean expressions, risk as a few percentages. The compiler validates everything; the runtime executes it.

Three things make ThunderScript different:

Quickstart

The fastest path to a running ThunderScript bot:

1. Open AiTrade and create a wallet (testnet recommended for first time)
2. Right-click the Bots panel → Create Bot
3. Pick a name + your wallet, set capital, click Create Bot
4. Click the new bot → ThunderScript Editor button (⚡ icon)
5. Replace the default with the example below, click Validate, then Save & Deploy
6. Press Start Bot on the bot card

Your first strategy

Here's a minimal but real strategy. It buys when EMA(9) crosses above EMA(21) with healthy momentum, and exits with a 1.5% trailing stop or RSI overbought:

strategy "Hello Momentum" {
  timeframe: 5m
  symbols: ["BTCUSDT"]

  indicators {
    rsi      = RSI(14)
    ema_fast = EMA(9)
    ema_slow = EMA(21)
    atr      = ATR(14)
  }

  entry when {
    ema_fast crosses_above ema_slow
    rsi in 35..65
    adx > 20
  }

  exit when {
    trailing_stop(callback: 1.5%)
    or rsi > 75
  }

  risk {
    max_position_pct: 5%
    stop_loss_atr: 1.5
    take_profit_atr: 4
  }
}

That's the entire strategy. The runtime fills in sensible defaults for everything else: AI Supervisor enabled, default model, no cooldown, etc.

Core concepts

Five mental models that make ThunderScript click:

1. The cycle loop

Every cycle_interval (default 5min), the engine runs your strategy once per symbol. It fetches indicators, checks conditions, and either opens a position, closes one, or waits.

2. Entry vs Exit logic

entry when uses AND logic — every condition must pass. exit when uses OR logic — any single condition triggers a close. This matches how humans actually think about trading: "I want all these to align before I buy, but I'll get out for any of these reasons."

3. AI as a final gate

If your strategy includes ai_context, the LLM is called only after all your conditions pass — as a final confidence check. This saves enormous LLM costs vs calling on every cycle.

4. Versioned forever

Every Save creates a new version with a performance snapshot. You never lose history. You can rollback, compare, or branch off any prior version.

5. Risk lives in the script

Stops, take-profits, position sizing, cooldowns — all in the risk block. The supervisor can tweak these alongside conditions; you keep one source of truth.

Syntax & Types

ThunderScript is whitespace-tolerant and uses braces for blocks. Comments start with //.

Strategy Block

Every script starts with a strategy block. The structure is fixed; sub-blocks are optional except entry when and exit when:

strategy "Name" {                  // required
  timeframe: 5m                       // optional, default 5m
  symbols: ["BTCUSDT"]                // or top_volume(N)

  indicators { ... }                  // optional
  entry when { ... }                  // REQUIRED — at least one condition
  exit when  { ... }                  // REQUIRED
  risk       { ... }                  // optional, sensible defaults
  ai_context { ... }                  // optional, LLM gate
  schedule   { ... }                  // optional, timing rules
}

Indicators

Declare what you measure. Each line creates a variable you can reference in conditions:

indicators {
  rsi      = RSI(14)
  ema_fast = EMA(9)
  ema_slow = EMA(21)
  atr      = ATR(14)
  vwap     = VWAP()
}

Conditions

Entry conditions (AND)

Every condition must evaluate to true:

entry when {
  ema_fast > ema_slow              // trend
  rsi in 35..65                   // not extreme
  adx > 25                        // directional
  price > vwap                    // above fair value
  not_in_cooldown(2h)             // no recent loss
  fee_ratio_ok(30%)               // fees won't kill profit
}

Exit conditions (OR)

Any single condition triggers a close:

exit when {
  trailing_stop(callback: 1.5%)
  or rsi > 75
  or ema_fast crosses_below ema_slow
}

Operators

Risk Management

risk {
  max_position_pct: 5%            // max 5% of capital per trade
  position_size_score_3: 25%       // score 3 → 25% of max
  position_size_score_4: 65%       // score 4 → 65% of max
  position_size_score_5: 100%      // score 5 → full max
  stop_loss_atr: 1.5              // SL at 1.5x ATR below entry
  take_profit_atr: 4               // TP at 4x ATR above entry
  cooldown_after_loss: 2h          // wait after a loss
  max_daily_loss: 5%               // stop trading if -5% today
}

Schedule

schedule {
  cycle_interval: 5m               // minimum 1m
  cautious_hours: [20,21,0,1]      // UTC hours: tighter rules
  cautious_confidence: 0.65        // vs 0.5 default
  cautious_adx: 30                  // vs 20 default
  cautious_size_penalty: 50%        // half size in cautious hours
  dead_hours: [3,4]                 // no entries (existing positions managed)
}

AI Context

ai_context {
  model: "deepseek"                // or "user.my-llm"
  prompt: "Trade only momentum setups."
  min_confidence: 0.5              // LLM gate threshold
  temperature: 0.3                  // LLM creativity
}

System models: deepseek, openai, anthropic, meta. User models prefixed with user.

Built-in Functions

Built-in Variables

Available without declaring:

Example: Conservative BTC

Holds spot BTC with strict trend + momentum filters. Big stops, slow cycles.

strategy "BTC Conservative" {
  timeframe: 1h
  symbols: ["BTCUSDT"]

  indicators {
    rsi      = RSI(14)
    ema_fast = EMA(21)
    ema_slow = EMA(50)
    atr      = ATR(14)
  }

  entry when {
    ema_fast > ema_slow
    rsi in 40..60
    adx > 30
    not_in_cooldown(4h)
  }

  exit when {
    trailing_stop(callback: 2%)
    or rsi > 80
  }

  risk {
    max_position_pct: 3%
    stop_loss_atr: 2
    take_profit_atr: 6
  }

  ai_context {
    model: "deepseek"
    prompt: "Only enter on strong weekly trends."
    min_confidence: 0.7
  }

  schedule {
    cycle_interval: 1h
    dead_hours: [2,3,4,5]
  }
}

Example: Aggressive Scalper

Hunts top-volume pairs on 5m. Many small trades, tight stops. Higher leverage allowed.

strategy "Scalp Machine" {
  timeframe: 5m
  symbols: top_volume(12, quote: "USDT")

  indicators {
    rsi      = RSI(7)
    ema_fast = EMA(5)
    ema_slow = EMA(13)
    atr      = ATR(7)
  }

  entry when {
    ema_fast > ema_slow
    rsi in 30..70
    adx > 20
    min_time_since_last_entry(15m)
    fee_ratio_ok(20%)
  }

  exit when {
    trailing_stop(callback: 0.8%)
    or rsi > 80
    or ema_fast crosses_below ema_slow
  }

  risk {
    max_position_pct: 8%
    stop_loss_atr: 1
    take_profit_atr: 3
  }

  schedule {
    cycle_interval: 5m
    cautious_hours: [0,1,2,22,23]
    cautious_size_penalty: 40%
  }
}

Example: Mean Reversion

Buys oversold, sells back to mean. Best in ranging markets (low ADX).

strategy "Range Hunter" {
  timeframe: 15m
  symbols: ["ETHUSDT", "SOLUSDT"]

  indicators {
    rsi  = RSI(14)
    bb   = Bollinger(20, 2)
    atr  = ATR(14)
  }

  entry when {
    rsi < 35
    price < bb.lower
    adx < 25                       // not trending
    fee_ratio_ok(25%)
  }

  exit when {
    rsi > 55
    or price > bb.upper
    or trailing_stop(callback: 1%)
  }

  risk {
    max_position_pct: 4%
    stop_loss_atr: 1.5
    take_profit_atr: 3
  }
}

Example: DCA + Grid

For non-ThunderScript modes, AiTrade also supports DCA (deterministic interval buys) and Grid (fixed-spacing rungs). Both are deterministic — no AI needed. Configure via the bot create form's Mode dropdown.

AI Supervisor

Every hour (configurable), the supervisor reviews each active bot's performance and optionally proposes a new ThunderScript version. It pulls:

• Last 24h trades, P&L, win rate, avg win/loss
• Per-symbol breakdown
• Last 5 versions' performance snapshots
• Top failing entry conditions
• Current market regime (volatile/trending/range)
• Your custom prompt (per-bot, max 500 chars)

It calls your bot's configured LLM with a structured prompt and parses a JSON response. If a valid ThunderScript is returned, it either auto-applies (default) or saves as a draft requiring your approval.

Auto-Revert

If a version loses >$500 with >5 trades in 24h, the engine auto-reverts to the best historical version. The new active version's baseline is reset, so subsequent checks measure from this point forward (no thrashing). A persistent notification + push alert explains exactly why.

Versioning

Every Save creates a new version. The runtime keeps a performance_snapshot attached:

Marketplace

Publish your strategies for the community. From the editor toolbar:

• Share — opens publish dialog. Title, description, tags, public toggle. Performance snapshot auto-attached.
• Export — downloads a .tscript file you can store anywhere.

From the Bots panel right-click menu: Marketplace. Browse by sort (Downloads / Upvotes / Recent / Performance), search, filter by tags. Each strategy:

• One-click Import → bot creation form pre-filled
• Toggle Upvote (community ranking signal)
• Copy Share Link → embed anywhere

Embed in your blog

Two ways to share strategies on external sites:

Clicking the link opens AiTrade with the import flow ready: review source, pick wallet, set capital, create bot.

MCP / API Tools

AiTrade exposes the supervisor's tools as HTTP endpoints (and over MCP for AI agent integration):

FAQ

How is my capital protected?

Multiple layers. Each strategy has stop-losses (ATR-based or fixed-%). Bot-level max_daily_loss halts trading after a bad day. Auto-revert pulls back losing versions. Pre-AI veto pipeline blocks trades in extreme volatility, low-edge setups, and book-thin markets.

Do I need an LLM API key?

Not necessarily. AiTrade ships with system providers. You can also bring your own (BYO-AI) — your key, your billing, but full control. Configure in My LLMs.

What if the supervisor proposes a bad strategy?

Three safeguards: 1) The proposed script must compile + validate (syntax & semantics) before being saved. 2) Auto-revert kicks in if performance drops. 3) You can disable auto-apply per bot — every proposal becomes an approval request you must review.

Can I run multiple ThunderScripts on the same bot?

No — one bot, one active script. But you can create multiple bots on the same wallet, each with its own script. They share capital but operate independently.

How do I migrate from another platform?

If your existing strategy is logic-based, translating to ThunderScript is usually 10-30 minutes. Indicators map directly. Conditions become entry when / exit when. Risk parameters fit the risk block. Reach out via hello@thundera.eu for help.

Is testnet supported?

Yes — Binance testnet for spot and futures. Paper bots use real prices but place no real orders. Recommended for first-time strategies.

Glossary

Changelog

v2.0 — May 2026

• AI Supervisor with approval workflow
• Strategy Marketplace + import/export
• Landing page
• iOS app v1.3+ (TradingView charts, ThunderScript editor, Live Activities)

v1.5 — April 2026

• ThunderScript DSL launched
• Auto-revert on performance degradation
• Tiered position sizing