Free 7-day trial · no card

Your EAs trade. AI Agents Supervisor protects.

You're running multiple trading robots. Good days they print money. Bad days they all lose together, and one bad day can kill the account. AI Agents Supervisor is the team lead watching every robot — pausing, scaling down, or closing positions before the losses pile up.

Works on MT5 with any EA — yours or bought from anyone.
See pricing See how it works
Works with any MT5 EA Bring your own LLM key Prop-firm aware
▸ Watches any MT5 robot · ▸ Works with any broker · ▸ OpenAI · Anthropic · Gemini · DeepSeek · ▸ Prop firm aware
Pricing · Free 7-day trial, then $299 for life

Run the full Supervisor free for 7 days — then keep it for life for $299.

Create a free account and bind one MT5 demo account in your dashboard — the full Supervisor runs with every safeguard for 7 days, no card. The clock starts when you bind, not at sign-up. Like it? One payment of $299 keeps it for life across up to 20 MT5 accounts — every v1.x update included, never a subscription.BYOK means LLM usage runs about $2–4/month for a typical setup.

Tier Free · 7 days · no card

Free 7-Day Trial The complete Supervisor — every safeguard, full Live mode — free for 7 days on one MT5 demo account.

$0 7 days · no card required
Sign up with email or Google, then bind one MT5 demo account in your dashboard — the 7-day clock starts at bind, not at sign-up. No card, nothing to cancel.
  • All 10 LLM agents — the full pipeline, end to end
  • Full Live mode — auto pause, scale-down and partial-close on your broker
  • One MT5 demo account bind — runs on your demo terminal
  • Full decision feed + JSONL audit log — every regime call, fit score, recommended action
  • Bring your own LLM key — OpenAI, Anthropic, Google, DeepSeek, xAI
  • Real-money (live) account — that needs Lifetime
  • Access past day 7 — that needs Lifetime
  • Up to 20 MT5 accounts · every v1.x update
Start your 7-day free trial
Free account, no card. Bind one MT5 demo account in your dashboard — the 7-day clock starts then.
★ Recommended
Tier Lifetime · One-time · v1.x updates

AI Agents Supervisor Lifetime Keep the full Supervisor for life — every safeguard, on up to 20 MT5 accounts, with every v1.x update.

$299 one-time · no subscription
Begins with the 7-day free trial
Begin with the 7-day free trial, then one payment of $299 keeps it for life — all 10 agents, every safeguard, every v1.x update, across up to 20 MT5 accounts.
  • Everything in the free trial — same pipeline, same audit, same prompts, kept for life
  • Live mode — Supervisor writes enabled · risk_mult · close_flag straight to your MT5
  • Cooperative + Adversarial dual control — Bull vs Bear vs Judge per cycle
  • Up to 20 MT5 accounts — bind your whole portfolio
  • Prop DD cap aware — FTMO, MyForexFunds, The Funded Trader presets included
  • Every v1.x update · lifetime · priority 1-on-1 Telegram support
Start the 7-day free trial, or pay once for lifetime now — no card needed for the trial. Across up to 20 MT5 accounts.

The 7-day free trial runs the full Supervisor — no card, nothing to cancel; the clock starts when you bind your MT5 account. Keep it for a one-time $299 (never a subscription) across up to 20 MT5 accounts. LLM cost example: at gpt-4o-mini pricing, watching 5 EAs on H1 ticks runs roughly $2–4/month in API calls — you control the model and the tick interval.

Ask AI · Inside MetaTrader 5

Just ask your account what's going on — and get a plain-English answer, right inside MT5.

The Supervisor isn't only a watchdog. There's a chat box docked right in your MetaTrader 5 terminal. Type a question — “How am I doing today?”, “Which robot should I trust?” — in any language, and the same AI engine that guards your account answers from live terminal data in seconds. No spreadsheets, no log files, no jargon. The conversation below is an illustrative example.

Ask anything about the account

Equity, balance, today's P&L, open exposure, daily and weekly drawdown versus your caps, and distance to your equity floor — answered from live terminal data, not a delayed report.

Know which robot to trust

“Which EA is performing best?” gets you a ranked answer — win rate, net P&L and profit factor — and flags the one quietly bleeding in this regime.

Get the next move, not just data

It doesn't stop at numbers. Ask what to do and it recommends a concrete risk action — scale down, pause, hold, or close — with the reason in one line. (It manages risk; it won't predict prices or give entry signals.)

Try asking
AI Agents SupervisorMT5 Live
How's my account doing today?
Supervisor
Equity$10,240
Today+1.8%
DD used1.2 / 5%
Up +1.8% today — two robots green, one flat. You're well inside your daily drawdown cap. Nothing needs your attention.
Which EA is performing best?
Supervisor
Gold Sniper leads at +4.2% this week (62% win rate). NAS Scalper is dragging at −1.1% in this chop — I'd scale it to 0.5× until volatility settles.
Should I worry about tonight's news?
Supervisor
High-impact USD print at 20:30. I'll auto-pause both USD robots 15 min before and resume once spreads normalize. You don't have to do anything.
Ask about your account…
New to trading robots? Read 30 seconds.

Three words. Then you're caught up. No jargon test.

01 — the software

MT5 — the trader's app

Software traders use to buy/sell forex, gold, indices. Think "brokerage app for global markets". The most popular one is MetaTrader 5.

02 — the robots

EAs — your robots

"Expert Advisors" — robots that place trades for you using preset rules. Traders run 3-5 robots at once to diversify. They mostly work; until they don't.

03 — what we add

Supervisor — the team lead

Sits above your robots, reads the market every hour, and tells the ones running the wrong play to pause / size down / close half. So one bad day stops being a fatal day.

Market turns ugly → 3 robots all losing → Supervisor pauses 2 → Account survives
Try it live

Three tools. Real math. No signup.

The same formulas the Supervisor runs at every cycle — wired to sliders you can poke. See if your setup survives, estimate AI cost on your usage pattern, and watch a regime switch play out across a basket of robots in real time.

Account survival simulator

Bad day hits a basket of robots. Does Supervisor save the account?

Live
Without SupervisorEvery robot trades fully
—
With SupervisorPauses 2 worst, scales 2 to 0.5×
—
Adjust the sliders to see the outcome.

BYOK monthly cost estimator

Pick model + check interval. We do the token math.

$3.40/month
2,880 AI calls/mo 8.6M tokens/mo $0.11/day

Billed by your provider directly. We never see your traffic or take a cut. Built-in budget cap auto-downgrades the model if daily spend exceeds your limit.

Regime → action demo

Pick the day's weather. Watch the basket react.

REGIME: Trending Bull confidence 0.87
Supervisor decision pipeline: snapshot → classify → score → command STEP 1 · EVERY TICK Snapshot Reads from MT5: • equity, margin • open positions • per-EA P&L 30d • news calendar STEP 2 · EVERY 15 MIN Classify regime LLM picks 1 of 5: 1 trending ↑ 2 trending ↓ 3 range 4 volatile 5 crisis + confidence STEP 3 · PER-EA Fit score [0–1] How well does EA's history match today's weather? ≥0.65 keep 0.40–0.65 scale <0.40 pause STEP 4 · ≤500ms Per-robot command Writes to: enabled · risk_mult dir_lock · max_sl_atr close_flag All to JSONL audit.
The Problem

Five robots that look safe alone will sink the account together.

Robots run the same markets, the same news, the same regime. When it turns, they don't lose one by one — they lose together. Here's the same $10,000 account, on the same bad day, without and with a Supervisor.

❌ Without Supervisor — high-volatility pullback day

Account: $10,000 · Daily loss cap: $500
Quant AI Agents XAUUSD −$80
London Breakout EURUSD −$120
Grid Recovery GBPUSD −$180
News Scalper USDJPY −$110
Trend Rider US30 −$60
Total day loss −$550
−$550 > $500 cap → prop account closed. Lost the −$550 + the $500 evaluation fee = −$1,050 in one day.

✅ With Supervisor — same day, supervised

Account: $10,000 · Daily loss cap: $500
Quant AI Agents RUN · full size −$80
London Breakout SCALE 0.5× −$60
Grid Recovery PAUSED $0
News Scalper CLOSE 50% −$50
Trend Rider RUN · full size −$60
Total day loss −$250
−$250 < $500 cap → account survives. Tomorrow the regime changes and the robots earn it back.

Numbers are an illustrative scenario, not a backtest. The Supervisor's job is to prevent simultaneous failures — not to eliminate losses.

How it works

Look. Read the day. Score the robots. Tell them what to do.

Every hour the Supervisor runs the same four-step loop. Cheap, auditable, boring on purpose.

STEP 01

Look

Read every open trade across every robot. Account size, position size, profit so far, room left before the cap. No AI yet — just inventory.

STEP 02

Read the day

What kind of day is this? Trending up? Range-bound? News-driven chop? An AI labels the regime with a confidence number.

STEP 03

Score each robot

Per robot: 0.0 to 1.0. "How well does this robot's playbook fit this kind of day?" Anything below your floor is on the chopping block.

STEP 04

Tell each robot

Five possible commands: RUN · SCALE 0.5× · PAUSE · CLOSE 50% · TIGHTEN STOP. Each carries a written reason. Rate-limited so it can't oscillate.

Two ways to take charge

Polite for our robots. Direct for everyone else's.

How the Supervisor stops a robot depends on whether the robot is willing to listen. Both modes ship in the box.

Mode 1 — Polite (our robots)

"Hey, stand down for a bit."

Our own robots (Quant AI family) check a shared message every second. The Supervisor writes "pause" / "scale to 0.5×" / "go full" — the robot reads it on its next tick and obeys. No order placed.

No broker cost. Nothing routes to the broker. No spread paid.
Surgical. Pause one robot while another keeps trading.
The robot stays the author. Supervisor only narrows when it can fire — doesn't override its logic.
Works on: Quant AI Agents · Quant AI Strategy · Quant AI Grid + every future sibling.
Mode 2 — Direct (everyone else's)

"Close half. Tighten the stop. Now."

Third-party robots can't be asked nicely — we don't have their source. So the Supervisor goes around them and uses MT5 itself to close part of the position and pull the stop in tighter. The robot wakes up to find its trade has shrunk.

Works on any robot. No source, no permission, no cooperation needed.
Two moves only. Close part of a trade. Tighten its stop. Never opens new trades the robot didn't open.
Cooldown. Max one move per robot per 30 minutes. Costs spread — enforced sparingly.
Works on: any robot from MQL5 Market, any custom build, anything inherited.
Earn trust by stages

Watch → Approve → Live. Nothing fires on day one.

The Supervisor ships in Watch mode. You see what it would do before any move is real — and you decide when it earns the trigger.

Stage 1 · default

Watch (Shadow)

Every hour: full decision + written reason logged. Nothing executes. You read the Supervisor's mind without paying for its mistakes.

The default. Live mode stays off until you turn it on.
Stage 2 · opt-in

Approve each move

Supervisor proposes. They wait in a queue. You click ✓ and the action fires. Train your gut on its instincts before trusting it solo.

Opt in whenever you're ready.
Stage 3 · live

Live

Moves fire automatically. Every move still has a reason in the log. One-click rollback if you disagree — and the Supervisor learns from the rollback.

Big actions still need the two-key rule.

Drop back to Watch any time. Promote one robot at a time. The Supervisor never auto-escalates above your current setting.

Inside the supervisor

Ten specialists. One audit trail.

Think of a small team — analyst, bull, bear, risk officer, portfolio manager. Each does one job. Every output is written down.

01

Technical Analyst

Reads the chart

Names what's on the screen: trend, range, breakout, reversal. The base everyone else builds on.

02

Multi-Analyst

Reads the basket

Watches all your robots as one portfolio. Flags "you're really making the same bet five times" traps.

03

Bull

"It keeps going up"

Argues the move continues. Lists the evidence. Names where it would be wrong.

04

Bear

"It reverses now"

Argues the move ends here. Lists exhaustion signals. Names where it would be wrong.

05

Risk Debate

The referee

How close are we to the loss cap? "Room for one more bad trade" or "no — flatten now."

06

Regime-Fit Scorer

Grades each robot

0.0 to 1.0 per robot. "Does this robot's playbook fit today?" The number that drives the call.

07

Correlation Reasoner

Spots stacked bets

3 robots long the US dollar in the same hour = 1 stacked bet. Penalises hidden concentration.

08

Supervisor PM

The decider

Reads everyone above and picks the command per robot. Writes the reason that ships in the audit log.

09

Intervention Agent

The hand

Actually moves. Sends pause / scale / close orders. Honours the cooldown so the Supervisor can't oscillate.

10

Reflection

Tomorrow's lesson

End of day: which calls were right? Which were wrong? What should we catch faster next time?

Every agent's full input + output sits in supervisor_log.txt in your MT5 data folder. Nothing hidden, nothing redacted, nothing sent to anyone but your own LLM provider.

Audit trail · Q&A

Every decision comes with the receipt.

Whenever the Supervisor pauses an EA or partial-closes a position, the decision lands in the panel with a rationale string. You can also ask follow-up questions inline.

Decision feed · last 5 actions LIVE
PAUSE Supervisor PM · 14:22:18: Grid Recovery Pro paused. Regime = high-vol pullback (conf 0.71). Grid strategies historically realise −2.3% in this regime; portfolio DD headroom is 1.2% — not enough buffer for a grid drawdown.
CLOSE 50% Intervention · 14:22:21: Closed 0.04 of 0.08 lots on News Scalper v3 USDJPY position #883121. Tightened SL to 156.92 from 157.40. Cooldown active for 30m on #720913.
SCALE 0.5× Supervisor PM · 14:22:25: London Breakout next-entry lot scaled to 0.5×. Regime fit 0.52 — not bad enough to pause, not good enough to size full. Restores to 1.0× when fit ≥ 0.65.
Q&A You · 14:23:01: "Why didn't you pause Trend Rider too? It's also USD-correlated."
Q&A Agent · 14:23:04: Trend Rider runs on US30 (index, not FX) and the Multi-Analyst flagged its correlation to the paused FX EAs as 0.31 — below the 0.5 stack-trap threshold. PnL today is also positive, so the regime-fit penalty hasn't crossed the pause floor (0.45). I'll re-score after the H1 close.
Risk safeguards

Defaults that fail safe, not loud.

An AI risk guardian that itself goes rogue is worse than no guardian. Nine guardrails ship on by default — every one of them tested under broken data, broken network, broken AI.

Shadow by default

New installs start read-only — the Supervisor watches and logs without touching your broker. Going Live is a deliberate one-toggle choice, and dropping back is just as easy.

Two-key rule for big actions

Close-all or any single move above 20% needs two AI calls to agree. One agent saying "do it" never fires.

If the AI is down → freeze, don't guess

API errors or junk responses N times in a row → lock new entries, keep existing trades on their own stops, alert you. Quiet beats wrong.

Cooldown per robot

Max one intervention per robot per 30 minutes (you set the number). No flip-flopping a robot on/off.

Heartbeat watchdog

Pings itself every 60s. Two missed pings → robots revert to their own defaults. Never stuck in "Supervisor said pause".

Full audit log on your disk

Every prompt, every response, every command — plain text in your MT5 folder. Forever. Yours to second-guess.

News calendar fails loudly

If your broker doesn't publish the high-impact calendar feed, you get a banner on every cycle until you fix it. The Supervisor never lulls you into thinking news is being checked when it isn't.

Multi-instance safe

Run one Supervisor per terminal, per account, per chart — every instance namespaces its state, files and decision logs by its own magic ID. No races, no overlapping commands, no two AIs fighting over the same robot.

Soft & hard license recovery

Brief network glitches don't lock you out — license check has a grace window, then falls back to read-only Shadow until connectivity returns. Your account stays protected even when our server can't be reached.

Built for

Three traders we built this for first.

Anyone running more than one EA benefits. These three are the people whose pain woke us up at 3am.

Audience 01

Prop firm traders

Funded by FTMO, MFF, TFT, FundedNext. One daily loss cap breach kills months of work.

"My account didn't die because one robot failed. It died because all five failed on the same news candle."

Audience 02

EA collectors

You own 3-5 robots. Some win in trending markets, some in ranges. You want them all on — just not all firing in the wrong kind of day.

"I don't want to choose one robot. I want them all on, supervised."

Audience 03

Multi-strategy hobbyists

You enjoy stacking robots from different authors and watching them duke it out. You don't want a babysitter — you want a referee that pulls them apart when the day turns ugly.

"I want my robots fighting each other for alpha. I just don't want them fighting me on a CPI candle."

A founder’s journey

One trader. A decade chasing the same question. What if the machine could think with me?

I’m Ho Tuan Thang — a solo quant who spent ten years in a small room in Chicago, building EAs no one asked for, blowing up demo accounts so live ones wouldn’t, and waiting for the technology to catch up to the idea. When large language models finally landed, the door opened. This is the short version of how we got here, and why I’m now spending most of my days helping other traders walk through it.

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

“I never wanted to sell signals. I wanted to build a co-pilot — something a serious trader could actually trust at 2 a.m. when the chart goes sideways. That’s the only thing I work on now.”

— Ho Tuan Thang · Founder, eatrading.ai
2014 · The start

Manual trader, blown account, rebuilt.

Started trading stocks and forex out of curiosity. Got humbled fast. Spent two years studying price action, risk management and the work of people like Van Tharp and Andreas Clenow — convinced the edge had to be systematic, not discretionary.

2017–2020 · First EAs

Solo dev, late nights, MetaTrader 5.

Learned MQL5 the hard way. Shipped a handful of rule-based EAs for friends, then a small client base. Every system traded my own money first. Lesson burned in: backtests lie, slippage is real, and an EA that needs a perfect market dies in a normal week.

2022–2024 · The AI turn

LLMs landed. The whole stack rewrote itself.

When GPT-4 and Claude opened up reasoning at machine speed, I rebuilt the engine from scratch. Started wiring multi-model ensembles into MT5 — real models, real keys, real accounts. Six iterations later, the Quant AI Agents framework was born.

2025–now · Helping the room

A small bench, a global crowd, one mission.

Today I split my week between shipping product and walking traders — in Vietnam, the Philippines, India, the EU and the US — through the messy migration from indicator EAs to AI-native systems. The goal is unchanged: arm the independent trader with the same firepower a hedge-fund desk takes for granted.

0+
Years in markets
0
EAs shipped to live accounts
$0M
Capital under management
0k+
Traders in the room
How the engine performs

The numbers behind every trade we ship.

Five LLM heads, millions of tokens, milliseconds of latency. Below is a snapshot of the AI machinery that argues out every order — engine telemetry, never marketing.

Ensemble
5/ 5 heads
LLM models in vote
Streaming
1.2M tokens / hr
Reasoned across cohort
Live
38.4k / day
Decisions emitted
Consensus
0.86avg
Ensemble confidence
Healthy
184ms p95
Inference latency
Closed-beta voices

What the beta cohort told us.

Quotes below are placeholder copy for the mockup. The live page will only use verified, opt-in quotes from traders running the AI Agents on live MT5 accounts — first name plus a masked role only, to keep prop-firm rules happy.

The debate transcript alone is worth it. I don't just see BUY — I see why three agents agreed and which one pushed back. Changed how I tune my own setups.
Marko K.
Prop trader · FTMO
I run 6 EAs across 3 brokers. Used to baby-sit them. AI Supervisor pauses two of them during NFP automatically — saved my account in March.
Diana J.
Retail · 4 yrs MT5
Free Quant AI Strategy was my entry. Ran it 3 months on demo, then bought the full Agents pack. The free version isn't crippled — it's just one agent.
Ryo S.
XAUUSD scalper
Prompt Library is gold. I tweak 2-3 lines, paste into the Lab, and the EA picks up a new behaviour overnight. It feels like having a quant team on tap.
Alex L.
Algo dev · MT5 builder
Most "AI EAs" are just MACD with a logo. This one actually changes its mind. I watched it skip a setup because CPI was 90 minutes away.
Sara P.
Indices swing
BYOK was the dealbreaker. I bring my own Anthropic key, see exactly what's spent, and nothing is locked behind a token meter the vendor controls.
Tomás N.
FX / XAU · 2 yrs MT5
Priority lane · partner brokers

Trade alongside the founder.

Open through any partner link below — same execution we run on live capital, plus a private lane reserved for traders in our community.

Beta seats for new EAs Direct line to the founder Tuned for AI workloads Priority 24/7 support
Built for EA workloads

IC Markets

Raw spreads, deep liquidity, execution tuned for heavy EA traffic.

Open account
Razor-thin commissions

Tickmill

Very fast fills on raw accounts — a strong second broker for spreading risk.

Open account
High leverage · EU

IC Trading

Higher leverage for EU traders who want capital efficiency on the IC backbone.

Open account

Affiliate / IB links — same broker conditions you'd get on your own, plus the community perks. None of your money goes to us.

Stay close

Three direct lines — walk into the AI era with the team.

Live trade updates on the channel, a 24/7 AI bot for product questions, and a direct line to me when you want a human. The market is moving fast; you shouldn’t have to wait days for an answer. Pick the channel that fits the moment.

Questions, briefly

FAQ

The short answers. Longer ones live in the docs.

Does it work with robots I bought from someone else?

Yes. That's Direct mode (Mode 2 above). The Supervisor uses MT5's own trade buttons to close part of the position and pull in the stop — it doesn't need the robot's source code or permission.

The robot doesn't even know the Supervisor exists. It looks like the broker just happened to fill a partial close. Most robots handle this fine; weird ones get flagged on day one.

What AI does it use? Do I have to pay extra?

You pick. OpenAI (we recommend gpt-4o-mini), Anthropic, Gemini, DeepSeek. You sign up at the provider, paste your key into our settings — your bill goes there, not us. We never see your traffic.

Cost is small: watching 5 robots once per hour on gpt-4o-mini ≈ $2-4 per month. Slower checks = cheaper. Premium model = more.

Will it work with my prop firm?

If the firm runs on MT5 and allows robots, yes. The Supervisor looks like just another robot to them. We ship presets for FTMO, MyForexFunds, The Funded Trader, FundedNext — for anyone else you type in the daily loss limit manually.

If the firm bans robots entirely (rare), the Supervisor is moot — and so is everything else you'd want to run.

How do I go from Watch to Live?

Three gates: (1) run Watch for 7 straight days, (2) open the dashboard and click "promote", (3) Live mode starts with a 24-hour cooldown before big-ticket actions unlock.

Going back to Watch is instant. Any move you rollback teaches the Supervisor — it adjusts the threshold the next day.

What if the AI provider goes down?

Circuit breaker trips. The Supervisor locks new robot entries, leaves existing trades running on their own stops, pings you on the panel + Telegram. It does not close trades aggressively just because the AI can't be reached.

When the provider's back, it self-tests two ticks in Watch mode before resuming Live — so a flapping API doesn't churn your account.

Can I read the source code?

The compiled robot file is what we ship. Source stays closed — the strategy is the product. But the full audit log is yours: every AI prompt, every response, every command, every rollback, in plain text on your disk. You can read every decision after the fact, without us in the loop.

Refund?

The 7-day trial is free and needs no card — see the full Supervisor work on your own account before paying a cent. After buying Lifetime: if your worst-day loss doesn't shrink versus the prior month, email us the audit log — refund. No hoops.

Will it work in the MT5 Strategy Tester?

No — by design. The Strategy Tester only runs one EA at a time, so there's no fleet to supervise. The Supervisor is built for live multi-EA terminals. To preview behaviour without risking money, run it in Shadow mode against your real demo or live account for a few days — that's the closest thing to a backtest for a portfolio-level tool.

Will it slow down MetaTrader?

No. Position-state checks run on a cheap 1-second timer. The expensive AI calls happen on a configurable cadence (default once per hour, never every tick). On a modest VPS you won't notice the Supervisor running next to 5+ EAs.

Is it safe to run on multiple charts or accounts?

Yes. Every instance namespaces its files, global variables, and decision logs by the Supervisor magic ID — no two instances stomp on each other. Run one per terminal, one per account, one per asset class. The cooperative gate that sibling EAs read also tolerates a stale Supervisor (network drop, restart) and fails open with a warning rather than locking your fleet.

What if my internet drops?

A configurable grace period covers brief outages. After that, the Supervisor blocks new entries on cooperative EAs and keeps existing positions managed safely on their own stops. Adversarial actions on third-party EAs are capped per hour, so a flapping connection can't spam partial-close orders into your broker.