Quant-AI

Speak in setups. Get a strategy.

Describe a trading idea in plain English. Quant-AI parses your intent, looks up live news, picks the right session, validates the spec, and queues a backtest — before it lands on your screen.

Avg compile
1.4 s
prompt → validated StrategySpec
Lines of code
0
no Pine, MQL, or Python
Schema
Zod-typed
every field validated server-side
First-shot pass
95%+
across 6 months of beta prompts
How it works

From sentence to signed-off spec, in four steps.

01
You describe

Plain English. Symbol, session, risk, RR, news rules — say what you want.

  • Voice or text
  • Reference past strategies
  • Template starters
02
AI plans

Quant-AI breaks the prompt into observable rules: indicators, patterns, sessions, exits, risk.

  • Looks up live news calendar
  • Picks timeframes by liquidity
  • Quotes broker hours and gaps
03
Spec compiles

A versioned StrategySpec v2 emits — JSON, schema-validated, tolerant of edge cases.

  • Zod runtime validation
  • Aliases for AI quirks
  • Method-agnostic (FVG, BOS, EMA, ATR, etc.)
04
Backtest runs

A 6-month backtest fires automatically. Replay arrives with KPIs, drawdown, and TP/SL ribbons.

  • No code path branches
  • Same engine as live
  • Margin-aware lot sizing
The spec

Self-describing. Audit-friendly. Versioned.

A StrategySpec is a single JSON object — every entry rule, exit rule, session, news guard, and risk cap is declarative.

One file. Every rule.
No black-boxes.

  • Indicators
    EMA, SMA, RSI, ATR, MACD, Bollinger, ADX, Stochastic, VWAP — composable per timeframe.
  • Patterns
    FVG, OB-reclaim, BOS, CHoCH, liquidity sweep, breakout, retest, engulfing, pin-bar, inside-bar.
  • Logic operators
    all/any/not nesting. Cross-timeframe. Cross-symbol. Cross-session.
  • Exits & risk
    ATR / pip / RR / swing stops. Breakeven (R or pips). ATR trailing. Margin-aware sizing.
  • Sessions & news
    IANA sessions, per-DOW, ForexFactory feed, regex kill-switches.
strategy.spec.json
{
  "version": 2,
  "name": "NAS100 EA v9 PA",
  "symbol": "NAS100",
  "entryTimeframe": "1m",
  "entryFill": "fvg_edge",
  "minFvgPips": 4,
  "entryLong": {
    "all": [
      { "htfTrend": { "tf": "5m",
                      "direction": "up",
                      "method": "ema_fast_gt_slow",
                      "fast": 20, "slow": 50 } },
      { "pattern":  { "kind": "bullish_fvg", "tf": "1m" } },
      { "levelBreak": { "tf": "1m",
                        "kind": "swing_high",
                        "lookback": 20 } }
    ]
  },
  "exit": {
    "stopLoss":  { "kind": "pips", "pips": 20 },
    "takeProfit": { "kind": "rr", "ratio": 1.2 }
  },
  "risk": {
    "perTradePct": 1,
    "leverage":    50,
    "dailyDDStopPct": 5,
    "stopOnConsecutiveLosses": 3
  },
  "sessions": [
    { "name": "NY-AM",
      "startUTC": "13:30",
      "endUTC":   "15:30" }
  ],
  "news": { "impact": ["high"], "avoidMinBefore": 15 }
}
What it understands

Twelve concept families. Hundreds of permutations.

Pure price action

FVG-edge entries, BOS / CHoCH structure, OB reclaims, liquidity sweeps, retests.

Indicator confluences

EMA stacks, RSI divergence, ATR breakouts, Bollinger squeezes, MACD crosses.

HTF / LTF alignment

Higher-timeframe trend filters with lower-timeframe entries — automatic.

Session logic

London / NY / Asia / overlap windows. Per-DOW. Per-DST behaviour.

Risk envelopes

Daily DD, max DD, per-trade %, consecutive-loss skip-day, leverage cap.

News & macro guards

ForexFactory red/medium/low folders, per-currency, regex kill-switches.

Cross-asset prompts

"Hedge XAU with DXY". The composer wires the relationship into one spec.

Version control

Every edit is a new spec version. Diff. Roll back. A/B side-by-side.

Backtest aware

When a prompt would fail (e.g. zero candles), the AI tells you why before running.

Inputs

Type, talk, or template.

Plain text

Your prompt, your words. The composer handles ambiguity by asking back.

Voice input

Hold the mic key and speak. Useful from the chart screen.

Templates

Cloneable starter prompts for ICT, SMC, ORB, NY-open scalps, news fades.

Guardrails

Validated before it ever runs.

Every spec passes through a Zod-typed schema. Anything the AI invents that the engine doesn't understand becomes a no-op (silent), never a crash.

Schema-validated

Field types, value ranges, enum membership — all enforced server-side.

Tolerant aliases

AI synonyms (e.g. ema_fast_lt_slow ↔ ema_fast_gt_slow + direction) normalised.

Unknown patterns

Patterns the engine doesn't recognise return false for that clause — strategy still runs.

Sanity checks

Stops within tick-size, RR > 0, leverage ≤ broker max, sessions within trading hours.

FAQ

Questions traders ask before trusting an AI composer.

Stop writing strategies in Pine.

Compose them in English. Audit the spec. Backtest before you deploy.

Join the waitlist