v1.0 A council of specialized AI agents on your MT5 chart

Five AI quant analysts debate every trade before it hits your account.

Analyst, Bull, Bear, Judge (Research Manager), and Decision (Portfolio Manager) — each running its own LLM call with calibrated confidence — analyse, debate, and only commit when consensus is reached.

Coming soon Contact me See how it works
Demo-first onboarding Bring your own API key Prop-firm friendly
Built for MT5 XAUUSD EURUSD GBPUSD NAS100 GER40 US500 M15 · H1 · H4 · D1 OpenAI · DeepSeek · Anthropic · OpenRouter · custom
The Pipeline

Five stages. One calibrated consensus.

Every signal travels the same path: Data → Analyst → Debate → Judge → Decision. No single agent gets to act alone — that's the entire point. If the Judge's confidence is LOW, the Decision agent downgrades to HOLD.

Stage 01

Data

OHLC across the chart TF and one above, indicators, prior swing levels, and the live MT5 economic calendar.

Stage 02

Analyst

Selects 3–8 most relevant indicators, identifies the regime, and flags one invalidating condition.

Stage 03

Debate

Bull and Bear each build evidence-based theses, address the opposing concern, name a failure mode, self-calibrate.

Stage 04

Judge · Research Manager

Synthesises the debate, runs a regime self-check, injects past lessons, commits to BUY / OVERWEIGHT / HOLD / UNDERWEIGHT / SELL.

Stage 05

Decision · Portfolio Manager

Final order construction with confidence + regime + RegimeAligned. Downgrades to HOLD if confidence is LOW.

On-Chart Panel

Read the reasoning, not just the verdict.

Every decision the agents make is shown in plain English on your chart. Three tabs: the analyst brief, the full bull-vs-bear debate, and the live order book. Nothing is hidden behind a black-box score.

Analysis tab

Regime Trend-up. Selected RSI(14) · ATR(14) · MACD · EMA-50 · Donchian-20.

Levels Support 2,038.40 · Resistance 2,058.10 · Invalidation H4 close < 2,033.00.

News NFP in 38h 12m. Trading halted starting T-2h.

XAUUSD · H4 · 14:32 GMT
Debate tab
Bull Trend-up confirmed. Conf HIGH.
Bear Sweep risk > 2,045. Conf MED.
Judge · RM Verdict OVERWEIGHT · 0.68
Decision · PM LONG · 0.05 · RegimeAligned
5 calls · 3.4s total · 2,210 tokens
Orders tab
XAUUSD LONG · 0.05
@ 2,042.50 · SL 2,035.20
+$18.40
EURUSD SHORT · 0.10
@ 1.0834 · BE moved
+$24.10
NAS100 LONG · 0.20
@ 18,210 · trail tight
−$6.80
Day P/L · +$35.70 across 3 charts
Input Management

Every input documented. Every default explained.

41 inputs across 4 groups — AI Engine, Prompt & Trading, Risk, Position Averaging. Defaults work out-of-the-box for XAUUSD H4. Most users only touch group 1 (paste API key) and group 2 (turn on Auto-Trade). Below: the AI + Auto BE/Trail safety net and the optional Position Averaging mode in detail.

Hybrid AI + Rule-Based Safety Net

AI + Auto BE/Trail (always on)

The Decision agent proposes the initial SL on every entry. From order fill, the local ATR engine takes over the stop — no API dependency, no rate-limit risk.

T+0 Order fills · AI-proposed SL accepted −1.0R
+1×ATR SL ratchets to break-even 0.0R
trail Trailing engages at 1.5 × ATR +ratchet
15s Engine ticks even if AI fails or rate-limits always
Optional · Risky

Position Averaging

Layered entry on adverse move with explicit caps. Per-position SL is stripped — the basket exits via group BE TP or kill switch only.

Max layers5
Layer spacing1.0 × ATR
Volume cap2.0 lots
Group kill switch−6.0%
Kill switch
Group DD breach closes the entire basket. No override, no model wait.

Need the full input reference?

Every one of the 41 inputs explained — type, default, when to change, and the watch-outs — in a dedicated reference page with search and difficulty filters.

View all 41 settings
Custom Prompts Library

Edit a text file. Append a strategy.

Strategies aren't hardcoded into the EA. They live in eatrading_prompts.txt at MQL5\Files\, in a structured [PRESET]…[/PRESET] format with NAME, SUMMARY, TIMEFRAME, WHITELIST, STRATEGY and RULES sections.

Custom presets append to the three builtins — they don't replace them. Cycle through all of them on the panel.

  • Auto-deployed template on first run with an example block.
  • One-click reload via the panel ↻ button — no EA restart.
  • Optional TIMEFRAME hint — advisory only; runs on any TF.
  • Reflection mode appends post-trade lessons back into context.
Founder

Built by a quant who actually trades.

I'm Ho Tuan Thang, founder and AI developer at eatrading.ai. I build the systems behind Quant AI Agents — every number below is conservative and verifiable.

Portrait of Ho Tuan Thang, founder of eatrading.ai
Ho Tuan Thang Founder · AI Dev
0
Years

In automated trading and applying Artificial Intelligence to live trading systems.

0
Years

Trading experience across stocks, crypto, and forex — including discretionary, manual trading.

$0
Capital under management

Total capital I manage across discretionary accounts for multiple investment funds.

0
Followers

Across the MQL5 channel, Telegram, X (Twitter), and other social platforms combined.

Trader Reviews

Trusted by retail and prop-firm traders.

4.8 / 5 · 240+ verified traders
Reading the Bull/Bear debate completely changed how I evaluate setups. The Judge calling out my own bias is worth the license alone — I skip far more bad trades now.
Marcus R.
Prop trader · FTMO
Connors RSI(2) preset on D1 gold has been astonishingly consistent. The kill-switch logic saved me twice during CPI surprise weeks. Honest tool, honest stats.
Amir K.
Retail · XAUUSD swing
I write my own prompts. The eatrading_prompts.txt format is dead simple and the panel reload button means I iterate in seconds, not minutes. Best workflow I've seen on MT5.
Julien T.
Quant dev · Paris
The Reflection layer is what sold me. Watching the agent learn from yesterday's losing NAS100 trade and refuse a similar setup today — that's institutional behaviour at retail cost.
Sara P.
Funded · MyForexFunds
Running on DeepSeek for thirty cents a month. The Auto BE/Trail every 15 seconds is why I sleep at night — even when the AI is rate-limited, my stops still ratchet.
Daniel H.
Retail · GBPUSD H4
Calibrated confidence labels (HIGH/MED/LOW) plus the Decision agent downgrading to HOLD on weak conviction — finally an EA that knows when not to trade.
Ricardo V.
Prop trader · The5ers
Pricing
Coming soon

Pricing announced at launch.

Two cost components: a one-time EA license fee, plus the AI provider's API fees billed directly by your provider. We don't mark up tokens — you pay your own usage.

~$2/mo per chart · GPT-4o-mini ~$0.30/mo per chart · DeepSeek BYO key OpenAI · DeepSeek · Anthropic · OpenRouter · custom
Join the waitlist See it run first
Priority customer · IB program

Register through my broker links — unlock the trader-community perks.

Sign up under one of the partner brokers below and you don't just get standard trading conditions — you get into a curated lane with direct access, beta seats, and 1-on-1 strategy support.

✨ When you register an account through my broker link, you receive standard trading conditions plus a set of privileges reserved for our trader community — built so you can actually compound on top of the AI agents instead of fighting fees, latency, and silent restrictions.

01 Optimize your trading results

  • Mirror the same execution conditions as my live signal accounts.
  • Beta-tester seat for new EAs and indicators — early access before public release.
  • Direct feedback channel to influence what ships in the next product cycle.

02 Superior support & personalization

  • Request EA customizations and tooling tailored to your account size and style.
  • Priority 1-on-1 consultation to build a trading plan that fits your goals.
  • Priority 24/7 support from the team for fast issue resolution.
  • Membership of a dedicated community with structured paths to develop your skills.
Engineered for EA

IC Markets

Raw spreads, deep liquidity, and execution profile tuned for high-frequency EA workloads. The default home for serious automated traders.

Open IC Markets
Raw spreads · ultra-fast

Tickmill

Razor-thin commissions on raw accounts and ultra-fast execution. A solid alternative when you want regulatory diversification.

Open Tickmill
High leverage · EU traders

IC Trading

Higher leverage tier for European traders who want capital efficiency without leaving the IC ecosystem. Same execution backbone.

Open IC Trading

🚀 Affiliate disclosure — these are partner / IB referral links. You pay nothing extra; the broker shares a portion of their spread with us, which funds ongoing EA development and the priority-support lane. Your trading conditions remain identical to (or better than) the broker's standard offering.

Stay connected

Join the channel. Or message me directly.

With a passion for technology and finance, I focus on building smart EAs — combining machine-learning algorithms, neural networks, and real-time data processing to optimize trading strategies and adapt to volatile markets.

Questions, asked honestly

FAQ

Which brokers and accounts work?
Anything that runs MetaTrader 5 with the standard symbols (XAUUSD, EURUSD, GBPUSD, NAS100, GER40, US500). The EA reads the live MT5 economic calendar, so calendar-equipped brokers get the news-aware Multi-Analyst behavior out of the box. Other brokers still trade — they just don't get the calendar pause.
Is it prop-firm compatible?
Yes. The Group Kill Switch DD, Floating DD Cap, and trade-frequency caps were designed against prop-firm rule sets. You set daily-loss and overall-loss thresholds in the consolidated Risk Management input group; the EA closes the basket when either fires. No martingale by default — Position Averaging is opt-in, gated by the same kill switch, and the size multiplier is set to 1.0 (flat) out of the box.
Do I need to provide my own API key?
Yes. We don't proxy LLM calls — the EA hits OpenAI, DeepSeek, Anthropic Claude, OpenRouter, or your custom endpoint directly using your key. You see your own usage, you control your own spend, and you can swap providers any time without contacting us.
What does it actually cost to run?
Per chart, on H4 with GPT-4o-mini: roughly $2/month in API fees. DeepSeek is the cheapest tested provider at ~$0.30/month per chart. Larger models (GPT-4o, Claude Sonnet) cost more. The EA caches identical contexts and only invokes the full 5-stage pipeline on signal candles, not every tick.
Should I run this on a live account immediately?
No. Run it on a demo account first — at minimum a full week of agent decisions on your broker, your symbols, your spreads. The agents adapt to whatever prompt you give them, and the only way to know whether your prompt and your broker pairing are sound is to watch the calibrated-confidence debate output for a few dozen trades.
Refunds?
14-day refund window for any reason — including "the agents make sense but I don't trust the results yet." We'd rather you test on demo, walk away, and come back when you're ready, than have you live with a tool you're not confident in.
Do the agents actually disagree, or do they all say "buy"?
They disagree often. The Bull and Bear must each address the strongest opposing concern and self-calibrate confidence as HIGH/MED/LOW. The Judge runs a regime self-check, and the Decision agent will downgrade to HOLD if confidence is LOW. Most of the value sits in the trades that don't happen — where the Bear's failure mode dismantles the Bull's setup. Watch a week of demo and you'll see this pattern.
Can I run on something other than H4?
Yes. Quant AI Agents is designed for H4 but runs on any chart period. Each preset declares its optimal timeframe (advisory hint to the AI) — H4 for swing, D1 for Connors RSI(2), but the engine will adapt the analysis to whatever you load. M15 / H1 / H4 / D1 are all supported.