Quant Mogul AI — Complete User Manual

What's inside

Three levels. Start at the top, or jump straight to what you need.

Part 1 · Getting Started

Up and running in five minutes

Everything a first-day user needs: install, connect your AI keys and MetaTrader 5, and let the app set up its own Python engine — no terminal, no dependencies, no stress.

1.First launch — the Setup Wizard

Appears automatically on first run · reopen any time via Help → Setup Wizard

The first time you open Quant Mogul AI, a five-step wizard walks you through the whole setup. Nothing you enter ever leaves your computer — keys are stored in the OS keyring on this machine.

Welcome aboard step — **Welcome aboard.** The wizard configures AI keys, connects MetaTrader 5, and installs the local analysis engine.

AI Keys — optional. The built-in Auto model works instantly on QuantMogul's cloud router using your WOBR credits — no key needed. Paste your own Anthropic (Claude) and/or OpenAI key to unlock every premium model at a lower credit burn (more providers — OpenRouter, DeepSeek, Kimi, Grok — live in Settings).
MetaTrader 5 — click Auto-detect. The wizard scans your machine and lists every installed MT5 terminal; click the one you want.
Local Engine — one click installs a managed Python engine (bundled interpreter + all libraries). This is what runs backtests, optimizations and deep-learning training locally.
Finish — a green checklist confirms everything is connected.

AI keys step — You can skip this step entirely — **Auto** needs no key. Any key you do enter is stored once in the OS keyring — masked on screen, never uploaded anywhere.

MT5 detection step — **Auto-detect in action:** the wizard found eight MT5 installations on this PC in about two seconds. One click selects the terminal.

Engine install step — Bundled Python 3.11 detected — press **Install engine (one click)** and the app does the rest.

Finish step — All three lights green: AI keys configured, MetaTrader 5 connected, local engine installed.

💡

TipSkipped a step? No problem. Every setting in the wizard also lives in the Settings page (section 3), so you can complete setup whenever you like.

2.Tour of the workspace — Build · Test · Deploy · Arena

This is home — you'll spend most of your time here

Quant Mogul AI looks and feels like a modern code editor, but every panel is built for trading-bot work. The center workspace is organized into four stages that mirror how a quant actually works — BUILD a strategy, TEST it against evidence, DEPLOY it live, and compete in the ARENA:

Main workspace — Left: workspace & tool rail (grouped BUILD / TEST / DEPLOY / ARENA) · Center: stage tabs with view pills below · Right: AI co-pilot · Bottom: terminal & live job logs.

Zone	What it does
Stage tabs	BUILD (Code, Node View, Strategy Map, Parameters, Bot Factory, DL Lab, MogulX) · TEST (Backtest, Replay, Optimization, Validation, 3D Lab, Data…) · DEPLOY (Live Run — Bot Farm) · ARENA (Global Arena, AI Marketplace, Alpha Engine).
View pills	The second row switches views inside the current stage — one click, no menu diving.
Left rail	Workspace files, Test Lab actions, Research tools, Data, Live Trading, Connections — grouped by the same stages.
AI co-pilot (right)	Chat with Agent / Plan / Ask modes. Writes MQL5, MQL4, Python and Pine Script — with the built-in Auto model or your own API keys.
Bottom panel	Terminal, Compile output, Backtest log, Optimization log and Problems — live-streamed from real jobs.
Status bar	Project name, active view, symbol/timeframe, MT5 connection, AI provider, latest strategy score.

Built for long sessions: every panel toggles instantly, and Focus mode hides everything except the center workspace with one shortcut:

Shortcut	Action
⌘B / Ctrl+B	Toggle left sidebar
⌘J / Ctrl+J	Toggle bottom terminal
⌘L / Ctrl+L	Toggle AI co-pilot panel
⌘⇧Enter	Focus mode (hide / restore everything)
⌘K / Ctrl+K	Command Palette — search any action or view by name

Code editor — Full Monaco editor (the same engine as VS Code) with MQL5 syntax highlighting, minimap and tabs. Note the starter prompts on the right — one-click AI recipes for your first bot.

3.Settings — keys, MT5 & engine

View pills → Settings (or the gear icon, bottom-left)

Every connection lives on one page. The essentials for a new user are the first three groups:

AI Providers — six key slots: Anthropic (Claude), OpenAI (GPT), OpenRouter, DeepSeek, Moonshot (Kimi) and xAI (Grok). All optional — Auto works with none of them. See Choose your AI brain below.
MetaTrader 5 — terminal path, MetaEditor path, data folder, and your demo account login / password / server for real Strategy-Tester runs.
Local Compute Engine — which Python runs your research jobs (set automatically by the wizard).

Settings page — This screenshot shows a real working setup: keys stored, MT5 demo account **109741389 @ MetaQuotes-Demo** connected, engine ready. Press **Save** (top-right) after any change.

🔒

SecuritySecret fields (API keys, MT5 password) are stored in the operating-system keyring — not in a plain-text file, and never uploaded anywhere.

3b.Choose your AI brain — Auto or your own keys

AI co-pilot → the model button next to Send (bottom-right)

The model selector sits right next to the Send button in the AI co-pilot. Every entry carries a badge that tells you exactly what it costs before you press anything:

Model picker with Auto, credit multipliers and locked models — **One click on the model button** opens the full menu: **Auto** on top (works with no key), your unlocked premium models with their ×N credit multipliers, and 🔒 locked entries for providers you haven't added a key for yet. Note the live credit balance in the title bar (⚡ 2450).

Badge	What it means
1–20⚡ (Auto)	Auto runs on QuantMogul's cloud router — no API key needed. It reads your request, picks the best-value model for the job and charges by task size: quick questions ≈ 1 credit, code generation ≈ 4, deep audits with huge context ≈ 20. You're only charged after a successful answer.
×N (e.g. Sonnet ×3, Fable ×10)	Credit multiplier when the model runs on your own API key. The provider bills you directly for tokens; QuantMogul burns only a small metering fee scaled by ×N — noticeably cheaper per request than Auto, and you get to pin exactly the model you want.
🔒 (locked)	No key for that provider yet. Clicking a locked model jumps straight to Settings → AI Providers so you can paste one.

Open Settings → AI Providers and paste a key for any provider you use — Anthropic, OpenAI, OpenRouter, DeepSeek, Moonshot (Kimi) or xAI (Grok). Press Save.
The matching models unlock instantly in the picker — the 🔒 disappears and the ×N badge shows your metering rate.
In Manage Models (right next to AI Providers), tick which models appear in the picker so the menu stays short and personal.

AI provider key fields and Manage Models list — Left: six provider key slots (each stored in the OS keyring). Right: **Manage Models** — tick the models you want in the chat picker. Add an OpenRouter key and press **Load models** to browse its catalog of hundreds of extra models.

✨

Zero-setup startJust subscribed? Pick Auto and start building immediately — no API account, no keys, no configuration. Your WOBR credits are the only thing it needs.

💡

Power-user moveAdd your own key for the models you love. You unlock the full premium lineup (Fable, Opus, Codex, Grok…), responses come straight from the provider with no relay hop, and the credit burn per request drops well below Auto's rate.

4.Build your first EA with AI

AI co-pilot (right panel) → pick a starter or type your own prompt

This is the moment Quant Mogul AI shows what it can do. We clicked the starter “Build a moving-average cross EA” — a single click, no typing — and watched the co-pilot work:

AI thinking — The Auto Router expands the one-click starter into a full spec (12/50 EMA cross, fixed SL/TP, everything as input parameters), routes it to the best model, and streams progress in the Agent Activity feed (bottom-right).

AI code applied to editor — **~30 seconds later:** the agent wrote **EMA_Crossover_EA.mq5 — 254 lines** — straight into the workspace: EMA crossover entries, fixed SL/TP, spread filter, one-trade-per-signal guard, all key values as input parameters. The file opens in the editor, ready to compile.

Review the explanation and the code preview in the chat.
The agent creates the file directly in your workspace (or delivers a reviewable patch with an Apply button when editing an existing file).
Press Ctrl S (or File → Save) if you make manual edits.

✨

Why this mattersYou just went from an idea to a complete, compilable Expert Advisor in under a minute — with Auto (no key required) or your own AI key, on your own machine. The co-pilot speaks MQL5, MQL4, Python and Pine Script.

5.Compile & run a real MT5 backtest

Left rail → Test Lab — the Trading Actions sidebar

The Test Lab sidebar is your mission control: a Run Setup block (symbol, timeframe, data source, date range, deposit) and a stack of one-click actions.

Choose Symbol (e.g. XAUUSD) and Timeframe (e.g. M15) in Run Setup.
Press Compile. For MQL5 files this invokes your real MetaEditor. Our AI-written EA compiled with 0 errors, 0 warnings in 490 ms.
Press Run Baseline Backtest. Quant Mogul launches the MT5 Strategy Tester headlessly, waits for the run, collects the report and parses every metric.

MT5 backtest log — The Backtest log (bottom panel) shows the real sequence: **Intaking EA → Launching MT5 Strategy Tester for XAUUSD M15 → report parsed**. Our test run executed **473 trades on 38 million real ticks** from the MetaQuotes demo server.

💡

TipMT5 backtests run on your broker's real tick history — the most accurate simulation possible. For faster iteration on Python bots, the built-in vectorized engine (next section) runs in seconds.

Part 2 · Intermediate

The research pipeline

Any pro will tell you: one good-looking backtest means nothing. This part walks the full evidence chain — backtest → optimize → walk-forward → robustness — on a real bot produced by the Bot Factory.

6.Python bots & the vectorized engine

AI co-pilot → language Python · or Bot Factory → Alpha Block Library

Python bots unlock the full research pipeline: multi-symbol portfolio backtests in seconds, optimization, walk-forward, robustness, Monte Carlo, deep-learning hybrids and live deployment. Ask the co-pilot for one, or let the Bot Factory manufacture them (section 18). For this walkthrough we use QF_AdaptiveMeanReversionWithVol705.py — a factory graduate that trades mean reversion with volatility bands and RSI confirmation.

🧠

What makes a bot “optimizable”?Python bots expose a PARAMS dict (lookback, entry threshold, RSI period…). Every number in it automatically becomes a dial that the optimizer, walk-forward and robustness engines can turn — no extra wiring.

7.Baseline backtest — the report card

Test Lab → Compile then Run Baseline Backtest · results in the Backtest Result view

Compile on a Python bot runs an engine check (imports, signal function, smoke run). Then the baseline backtest runs on the vectorized portfolio engine across five symbols at once — forex, gold, crypto and indices in a single run:

Python backtest report — The report card: **+0.7% return, Sharpe 6.77, max drawdown −0.5%, 68 trades, 69% win rate** across EURUSD, XAUUSD, GBPUSD, BTCUSDT and NAS100 on M15 — with the full equity curve and trade markers below. One click, about a minute.

Equity curve — The dedicated Equity Curve view adds CAGR and Calmar, and shades every drawdown period. From here one click takes you to **Visual Replay** to watch these exact trades bar by bar.

📉

Honest numbers onlyQuant Mogul never massages results. A strong baseline like this one still has to survive optimization, walk-forward and Monte Carlo before it earns real money — and you'll see below that the pipeline catches exactly the weaknesses a pretty backtest hides.

8.Parameter optimization

Test Lab → Run Coarse Optimize then Run Fine Optimize

Coarse Optimize grid-searches the bot's parameter space with an in-sample / out-of-sample split. Fine Optimize then zooms into the best region with a tighter grid.

Optimization run — Every pass in a sortable table: swept ranges up top, then **SHARPE next to OOS_SHARPE** for each combination — the two columns that expose overfitting at a glance. Best pass here: in-sample Sharpe 14.5, out-of-sample 10.1.

🎯

IS vs OOSEvery optimization holds out data the optimizer never sees. If in-sample looks brilliant and out-of-sample collapses, the parameters memorized the past instead of learning the market. Walk-forward (next) makes this test systematic.

9.Walk-forward & the WF Matrix

Test Lab → Run Walk Forward · then Run WF Matrix

Walk-forward slides an optimize-then-trade window across history: optimize on window A, trade the result on unseen window B, roll forward, repeat. The efficiency ratio (out-of-sample performance ÷ in-sample performance) is the single most honest number in strategy research.

Efficiency	Meaning
> 0.6	Excellent — the edge survives on unseen data.
0.3 – 0.6	Usable — proceed, but validate further.
< 0.3	FAIL — the strategy is fitted to the past. Rework it.

A single walk-forward run can get lucky with its fold boundaries. The WF Matrix removes that luck: it re-runs the whole walk-forward across 12 configurations (fold counts × out-of-sample shares) and colors each cell green (passed) or red (failed):

Walk-forward matrix — **Caught red-handed:** the optimizer's “best” parameters from section 8 fail every walk-forward configuration — the matrix stamps them **OVERFIT · 0% pass**. The baseline parameters were fine; the “improved” ones memorized the past. This one screen just saved a live account.

🛡

The QuantMogul differenceMost retail platforms stop at the pretty backtest. Quant Mogul is built around the uncomfortable second question: “will it survive data it has never seen?” — and asks it twelve different ways before you risk a cent.

10.Robustness & the 3D Lab

Test Lab → Run Robustness Test · then View pills → 3D Lab → Robustness Cloud

The robustness test perturbs your bot's parameters hundreds of times and re-runs each variant. A robust strategy keeps making money when its knobs are slightly wrong — a fragile one falls apart. The result feeds a live 3D visualization:

3D robustness cloud — **200 real parameter samples** rendered as a rotatable 3D cloud — each dot is one variant, height = final equity. Drag to orbit, scroll to zoom. Our bot scores **Robustness 95.6/100 with a 0.0% ruin rate** — the tight green cluster above the disc means the profitable region is a plateau, not a needle. The scene carries a **LIVE data** badge because it's your actual run, not a canned demo.

The 3D Lab has six more scenes — Optimization Surface, Equity Curve 3D, Walk-Forward, Portfolio, Stress Fan — each one wired to your latest real results.

11.Trade list & visual replay

View pills → Trade List and Visual Replay

12.Node View, Strategy Map & Parameters

View pills → Node View / Strategy Map / Parameters

Three different lenses on the same strategy:

Node view exact — Bots born from the Alpha Block Library carry a **QM:LINEAGE** record, so Node View rebuilds their logic **one-to-one** — see the **EXACT — from QM:LINEAGE** badge. Price → EMA cross entry → vote → signal, exactly as the code executes. Not a drawing; the actual logic.

Node view approximate — For hand-written or AI-written bots without lineage, the structure parser draws an **APPROXIMATE** graph from the code itself — here it detected SMA, RSI and Bollinger indicator blocks feeding the entry logic. You can also start a **new editable graph** from a template, wire nodes and press **Apply → Code View** to generate MQL5, MQL4, Python or Pine.

Strategy map — The app parses your bot and maps its anatomy: Market Data, Signal Engine, Risk Engine, Order Engine, Exit Engine and Filters — all green for this bot — plus a trading-logic checklist below (entry, exit, indicators, long+short, position sizing detected; stop loss and take profit flagged as missing). A to-do list for hardening your bot.

Parameters grid — All 5 inputs parsed from the bot with auto-suggested start/stop/step ranges. Tick the params you want optimized — this grid is exactly what Coarse/Fine Optimize sweeps.

13.Exporting reports & set files

Test Lab → Export Report / Save Best Set File / Compare Results

Export Report — writes a Markdown research report (setup, metrics, robustness numbers) into your workspace and opens it in a tab. Perfect for a research journal or sharing with a partner.
Save Best Set File — after an optimization, saves the winning parameters as a .set file (MQL5 bots, loadable straight into MT5's Strategy Tester) or params JSON (Python bots).
Compare Results — side-by-side table of every run in this session, so you can prove a change actually helped.

Exported report — Research report generated and opened automatically — strategy, run setup, robustness runs and ruin rate, timestamped.

Part 3 · Advanced

The professional arsenal

Institutional-grade validation, autonomous strategy mining, local deep learning, multi-broker bot farms and a worldwide arena — the tools that separate a hobby from an operation.

14.Monte Carlo Validation Lab

View pills → Validation Lab · or Test Lab → Run Monte Carlo

The question that matters is not “did it make money in the backtest?” but “how likely does this survive the real world?” The Validation Lab answers it with an institutional battery: order-shuffle and block-bootstrap Monte Carlo, execution-friction simulation (slippage, spread, missed fills), parameter sensitivity, walk-forward Monte Carlo, market-regime stress, risk of ruin, and tail metrics — condensed into one Robustness Score.

Monte Carlo fan chart — The fan chart shows the median path and 5–95% envelope across 1,000 reshuffled trade orderings against the hard-stop line, with the max-drawdown distribution below. The per-category bars on top (Bootstrap, Execution, Parameter, Walk Forward, Market Regime, Risk) show exactly which dimension failed — here parameter sensitivity and walk-forward dragged the score down, echoing the WF Matrix verdict from section 9.

Findings and action plan — Plain-language findings with a concrete action plan for each failed category. There's even a one-click prompt to have the AI explain the results and propose fixes.

🏛

PASS / CAUTION / FAILThe lab applies a hard rule set (≤5% ruin probability passes, 5–10% cautions, >10% fails) so the verdict is objective — the same discipline hedge funds apply before allocating capital. Every gate exists to kill bad bots while they're still free.

15.Strategy Discovery Engine

Left rail → Research → Discovery

Don't have a strategy idea? Evolve one. The Discovery engine breeds populations of rule-based strategies on the vectorized engine — thousands of candidates per second — with a built-in out-of-sample holdout so overfit candidates are penalized before you ever see them.

16.QuantData Manager

Stage TEST → QuantData Manager

Research is only as good as its data. The QuantData Manager pulls bars from any connected broker (MT5, Binance, Alpaca, CCXT, Databento), free institutional Dukascopy tick data, or Yahoo — then cleans, gap-checks and quality-scores every dataset and caches it as Parquet. Every backtest, optimization and DL training run draws from this store automatically.

17.Analyzer Suite

View pills → Analyzer

One good bot is a start; a portfolio of uncorrelated bots is a business. The Analyzer blends strategies, computes correlations and combined stats, runs portfolio-level Monte Carlo with probability of ruin, and sweeps money-management settings.

Analyzer suite — Portfolio analysis with equal-weight / momentum blends and a money-management sweep. Below it, the **live-vs-backtest drift monitor** keeps watching after deployment and raises the alarm when reality decouples from the backtest.

18.Bot Factory & the Alpha Block Library

Stage BUILD → Bot Factory

The most autonomous tool in the app — and it doesn't need to burn a single AI credit. The Alpha Block Library is a curated catalog of trading concepts (trend following, mean reversion, breakout, momentum, price action, multi-symbol, calendar, machine learning…), each with concrete plays: EMA fast/slow cross, price bounce off EMA, channel breakout, RSI reversion and many more.

Pick a preset chip (e.g. “Trend starter”, “Breakout pack”) or tick individual plays.
Set profit criteria and budgets, press Start. Every candidate must survive backtest → walk-forward → robustness before it reaches you.
Graduates land in your workspace as .py + .mq5 with a QM:LINEAGE record — so Node View shows their exact logic graph, and the Arena knows their ancestry.

🧬

Learning memoryFailures feed a learning memory, so each generation of candidates gets smarter — the factory literally evolves its own search as it runs.

19.Deep Learning Lab — train on your machine

Stage BUILD → DL Lab

Train a real neural network on your own machine — your data, your CPU, your model. No cloud bill, no notebook, no PhD. Metrics are honest out-of-time numbers (the validation tail is never trained on), and one click composes the trained model into a runnable Python bot with the weights embedded.

DL Lab RL policy — Pick a brain: **Forecaster** (predicts next-bar returns), **Pattern Net** (temporal window network reading lagged market history), or **RL Policy** — reinforcement learning trained with evolution strategies, rewarded on **profit after costs**, that decides positions directly. Training streams a live loss/reward chart, and **Continue training** warm-starts any model on fresh data after a live period.

🔒

Your alpha stays homeOnly the scorecard (metrics) can be shared to the global ranking — never your weights, code or data.

20.MogulX DL Trader — 20 architectures

Stage BUILD → MogulX DL Trader

Want the heavy artillery? MogulX ships 20 deep-learning architectures — LSTM, Temporal Fusion Transformer, PatchTST, N-BEATS, Informer, TimesNet and more — with a full training pipeline and an MT4/MT5 bridge EA that trades the signals straight from MetaTrader.

21.Live Run — deploy a bot farm

Stage DEPLOY → Live Run · Left rail → Live Trading and Connections

The payoff. Your Python bot runs on QuantMogul: it pulls fresh bars from a data venue, computes signals locally, and fires orders out to MT5 (native or file-bridge), MT4, IBKR, Alpaca, Binance, CCXT or the paper simulator — with risk-based position sizing, stop-loss / take-profit / trailing management and a kill switch. And you don't run one bot — you run a farm:

Live run bot farm — The farm sidebar (left) launches any number of bots — each picks its own strategy file, symbols, data venue, execution venue, risk %, SL/TP/trailing and check interval, and runs independently. The dashboard tracks farm bots, iterations, equity and open positions across all of them. Note the **Arm live orders (real money)** checkbox — unchecked by default.

Everything starts in dry-run: the full loop runs and logs every decision, but no orders leave the machine.
Watch the iteration counter, equity, positions and pending-order tables update live. Ticks stream back from MT5 for the bot's calculations; every fill is logged for the stats.
Only when you're satisfied do you arm live trading — an explicit confirmation step. The kill switch flattens everything instantly.

Connections panel — The Connections rail manages your primary MT5 plus **extra terminals**: register every additional MT4/MT5 terminal on the machine (one per account), and point each bot at whichever terminal it should trade into. Multiple bots per terminal, multiple terminals per machine — a real farm on one computer.

⚠

Safety by designDry-run is the default everywhere. Real-money trading requires explicitly arming the runner, and MT5 one-click launch attaches the bridge EA to your demo account first. Never arm a bot that hasn't passed the Validation Lab.

22.Global Arena, DL Marketplace & Alpha Engine

Stage ARENA → Global Arena / AI Marketplace / Alpha Engine

When your bot graduates, let it compete. The Global Arena is a worldwide leaderboard of QuantMogul builders — publish your bot's scorecard (metrics, symbols, lineage — never the code, weights or data) and battle other bots head-to-head, metric by metric.

Marketplace — Ready-made ML bots: one click composes the full Python source into your workspace so you can verify it on **your** data before trusting it — then backtest, optimize and live-run it like any other bot. Hosted signal models and a native MT5 deep-learning indicator template round out the offering.

Alpha engine — Every client can stream anonymized backtest / optimization / walk-forward telemetry into the central engine, which distills millions of runs into a global **alpha score**, a strategy leaderboard and crowd-sourced parameter ranges fed back into your optimizer. Connect the Signal API in Settings to join the network.

23.Workflow cheat sheet

The proven path from idea to live bot farm:

Step	Tool	Gate to pass
1 · Create	AI co-pilot / Node View / Bot Factory / DL Lab / Discovery	Compiles clean (engine check or MetaEditor 0 errors)
2 · Baseline	Run Baseline Backtest	Positive expectancy, enough trades to be meaningful
3 · Optimize	Coarse → Fine Optimize	Out-of-sample result must confirm in-sample — watch the gap
4 · Walk-forward	Run Walk Forward + WF Matrix	Efficiency ≥ 0.3 and a green matrix — not one lucky fold
5 · Robustness	Robustness Test + 3D Cloud	Profitable region is a plateau, not a needle
6 · Validate	Monte Carlo Validation Lab	PASS verdict, ruin probability ≤ 5%
7 · Sharpen	DL Lab — confirmation gate / RL policy	Hybrid beats the pure-indicator baseline out-of-time
8 · Blend	Analyzer Suite	Low correlation to your existing bots
9 · Deploy	Live Run farm (dry-run → arm)	Dry-run behaves exactly like the backtest
10 · Monitor	Analyzer drift monitor	Live equity tracks the backtest envelope
11 · Compete	Global Arena	Publish the scorecard — never the code

🏁

Golden ruleEvery gate exists to kill bad bots cheaply. The earlier a weak strategy dies, the less it costs you. Quant Mogul AI makes each gate one click — so you'll actually use them.