Your robots trade. The desk watches every one.
Hard limits are live across your whole MetaTrader 5 terminal — your own EAs and any third-party EA alike: a risk-per-trade ceiling, a daily loss cap counting open and closed trades, position limits, and a kill switch that stops new trades before they reach your terminal. The AI layer above them — reading the live regime, scoring how well each robot fits it, and answering you in plain language — is in development. Both obey one rule, and it is enforced in code today: it can only ever turn risk down — it never opens or enlarges a position.
Every EA on your account, under one set of limits.
The limits are live: every open trade across every robot counts against one daily, weekly and monthly loss cap, one equity floor and one position ceiling, enforced on the server before an instruction reaches your terminal. Per-robot regime scoring — the Fit and Action columns below — is in development; the fleet shown here is illustrative, not a reading of your account.
| Robot | Symbol | Owner | Fit | Action |
|---|---|---|---|---|
| Gold EA | XAUUSD | yours | 0.81 | RUN |
| Gold Scalp | XAUUSD | yours | 0.55 | SCALE 0.5× |
| London Breakout | EURUSD | 3rd-party | 0.30 | PAUSE |
| Grid Recovery | GBPUSD | 3rd-party | 0.27 | CLOSE 50% |
| News Scalper | USDJPY | 3rd-party | 0.69 | RUN |
The AI proposes. The governor guarantees.
A deterministic safety layer that needs no AI sits under an agent debate that adds nuance on top. Even if every model is wrong or offline, the supervisor can only ever turn risk down.
Deterministic hard governor
A daily loss cap that counts open and closed trades, weekly and monthly caps, a hard equity floor, a risk-per-trade ceiling and position limits — enforced in code, on the server, before an instruction can reach your terminal, and no limit can be relaxed from inside a lockout. It runs with the AI completely offline. This is the floor nothing falls through. Weekly and monthly caps, a hard equity floor, a margin guard that stops you above your broker's stop-out, and a news blackout that pauses entries around high-impact releases are live too — each off until you set it.
Exit engine
Once a position is open, three exits can manage it for you: move the stop to break-even after the trade has made what it originally risked, move it to break-even after a set time if the trade is in profit, and close part of the position at a set R-multiple. Each is off until you set it. They apply to every position the terminal reports, including ones Trade X did not open. They run each time your terminal reports, not on every tick.
Break-even moves the stop to entry and never past it — moving a stop into profit is the trail's job, not this. R is measured against the stop the trade opened with, not against a stop the trail has since moved. An ATR-aware trail and a volatility-spike emergency flat are in development; what ships today is a percent-of-price trail.
Seven-voice agent debate
Not running yet. The design: Technical, Bull, Bear, Correlation and Regime analysts feeding a Research Manager and a Portfolio Manager, which would propose reductions only — the deterministic governor beside this card already refuses anything that raises risk, and would refuse these the same way. Nothing on your account is steered by it today.
Look at the whole account. Reduce only.
Steps 01 and 04 run today and are enforced on the server. Steps 02 and 03 — reading the regime and scoring each robot against it — are in development, and are marked as such below rather than described as though they were running.
Look
Read every open trade across every robot — exposure, P&L, room to your cap. No AI yet, just an honest inventory of the account.
Read the day in development
The agent debate would label the live regime — trend, range, volatile, crisis — with a confidence, so action is anchored to conditions rather than a hunch. Not built yet.
Score the robots in development
Each robot would get a fit score for the current regime — a grid system in a crisis scoring low, a trend EA in a clean trend scoring high. Not built yet: today the limits apply to the account as a whole, not per robot.
Reduce only
Pause, scale down, tighten the stop, lock a direction or partially close — whichever lowers risk. It never opens or enlarges a position.
Does it save the account on a bad day?
A bad day hits a basket of robots. Move the sliders and watch what the loss looks like with the supervisor versus without it. The figures are plain arithmetic on your inputs — illustrative, and exactly the kind of maths the governor runs locally.
Pick the day's weather. Watch the basket react.
It can only ever turn risk down.
The only actions in the whole system are pause, scale down, tighten a stop, lock a direction and close. There is no code path that opens or enlarges a trade — the reduce-only promise is enforced in code, not just stated in copy. Your own EAs cooperate gently through a one-line include; third-party EAs you don't own are handled broker-side and need no changes.