Why AI Trading Platform

Not another indicator, but a complete infrastructure for autonomous algorithmic trading

It runs without you

Bots and training live on the server. Close the browser, switch the computer off — trading and model training carry on around the clock.

Four brokers in one window

Binance, OANDA, IG and Trading 212 — crypto, forex, indices, commodities and equities from a single terminal with shared statistics.

Transparent mathematics

No black boxes: win rate, average win and loss, and expectancy for every bot. Negative expectancy is filtered out.

Capital protection

A multi-level MarginGuard, a budget limit per bot, and paper mode by default for every experimental strategy.

Flexible configuration

Lot sizes, confidence, timeframes, instruments, target and stop in dollars or pips — everything adjusts to your style and risk.

Backtesting and shadow bots

Strategies are validated on two years of history (walk-forward) and run in shadow mode with no risk before going live.

Profit protection ratchet

The day's gain is locked in steps: give back 10% from the peak and that level becomes a floor which is never surrendered. Profit only moves upward.

Weekend protection

Before the market closes, profitable positions are booked and losing ones get a tight stop (≤ £0.10) — protection against weekend gaps and swap charges.

Capital tiers

As the account grows, the number of simultaneous trades and strategies grows rather than the size of each — risk per trade stays constant.

Kill switch and daily stop

A guardian daemon watches net asset value and shuts trading down on an abnormal drawdown; each bot also has a daily loss limit. Protected against false triggers from data failures.

Daily capital stop

A hard limit: once the day is down beyond the threshold, every position closes and there is no more trading until tomorrow. The maximum daily loss is capped and capital comes first.

Email alerts

A health monitor every 5 minutes: a bot down, a spike in errors, a silent failure, low disk space — an immediate email, deduplicated, with a follow-up when it recovers.

Recovery snapshots

A daily archive of code, configuration, bot state and AI weights. After a major failure the system is restored from the latest snapshot with a single command.

Every experimental bot runs in paper mode by default — real money is only involved deliberately, by an explicit switch to LIVE.

DEMO and LIVE trading

Two operating modes and layered protection against a margin call

DEMO mode

A starting balance of $10,000, saved locally, with the wallet resettable at any time. AI weights for every bot persist between sessions.

LIVE (OANDA v20)

Real trades on OANDA, synchronised every 10 seconds, with local IDs mapped to OANDA trade IDs. Margin: forex 1:30, indices 1:20, commodities 1:10.

MarginGuard

Warning at 200% → Danger at 150% (new trades blocked) → Closeout at 120% (worst position closed automatically) → Emergency at 100% (all positions closed).

Security and access protection

Layered protection: network · application · data

Bcrypt

Passwords are bcrypt hashes (salted, deliberately slow), unreachable from the web and resistant to brute force.

CSRF + Session

A 64-character token in every form, with the session ID regenerated on login and every 5 minutes.

Anti-brute-force

Five attempts per IP address in 15 minutes, with the session bound to the IP and user agent.

CAPTCHA

An image CAPTCHA on the login page turns away bots and automated password guessing.

WAF (ModSecurity)

The OWASP Core Rule Set blocks SQL injection, cross-site scripting and path traversal before they reach the application.

Fail2Ban

Automatic IP bans on attacks, scanning and brute force — enforced at the firewall.

Secrets are locked away

API keys, .env files, sources and logs are unreachable over HTTP — they return 403.

Cookies and headers

Secure · HttpOnly · SameSite=Strict, plus HSTS, CSP and X-Frame-Options.

In addition: rate limiting on every action (protection against flooding and runaway requests), an audit log of logins, one-shot orders through server-side daemons, and paper mode by default for new strategies. No file uploads and no execution of external code — the usual web-worm vectors are closed.
Total codebase — ~14,700+ lines: 23 JavaScript modules on the front end, a PHP API and a Python training daemon on the server. Version v20260310b · client DevIT.
×