Open a position on most “AI trading” apps and you’re really trusting a single model’s read of the market, delivered with more confidence than the data usually supports. ZAINEX takes a different path. Underneath its dark, ticker-lined interface sits a deterministic technical engine — the same trend, momentum, volatility, and support/resistance math a discretionary trader would run by hand — and only after that snapshot is locked in does an AI model get a look at it, constrained to a strict BUY / SELL / WAIT vocabulary it isn’t allowed to argue its way out of.
The platform calls this layer InteliBrain, and it is deliberately boring in the way safety-critical software should be boring. The technical snapshot — EMA9/21, RSI14, MACD histogram, ATR14, relative volume, support and resistance — is computed first, in code, with no model involved. A backend-enforced “safety gate” then decides which recommendations are even legal for that snapshot: if the deterministic score doesn’t clear a threshold, the AI’s only legal answer is WAIT, full stop, no matter how persuasive its reasoning sounds.
01Markets, not just a market
ZAINEX runs three asset classes under one interface: crypto, forex, and stocks. Crypto spot and futures currently span seven pairs — BTC, ETH, SOL, BNB, XRP, ADA, and DOGE — priced from live Binance data with automatic failover to alternate exchanges if one feed is unreachable. Forex just grew to eight pairs with the addition of XAU/USD, sitting alongside EUR/USD, GBP/USD, USD/JPY, AUD/USD, USD/CAD, USD/CHF, and NZD/USD. Stocks round things out with delayed daily data.
02Eight opinions, one deterministic referee
Today, live signal generation runs on GPT-5.1. But the architecture was built to be model-agnostic from the start, and ZAINEX is in the process of rolling out seven more: Claude Sonnet 4.5, Gemini 3 Pro, Grok 4.20, Grok 4, DeepSeek Chat V3.1, Qwen3-Max, and Kimi K2 Thinking. The idea isn’t novelty for its own sake — different model families trained on different data tend to disagree at the margins, and a signal that survives several independent reads is a more interesting signal than one that only ever heard itself think.
“WAIT is always an allowed answer, and it’s the default whenever the signals disagree.”
Every recommendation still ships with the receipts: the technical score that produced it, the risk level, entry/stop-loss/take-profit levels, and a short list of reasons a human can actually evaluate — not a black box, a worked problem.
03Guardrails, not autopilot
The most consequential design decision in ZAINEX may be the least visible one: the AI never places a trade. Every signal response is stamped autoExecute: falseat the API level, and every order — spot or futures — requires an explicit user confirmation. Futures positions carry configurable leverage up to 20x with isolated margin and a simplified liquidation model; stop-loss and take-profit levels, once set, are enforced automatically so a plan made with a clear head survives a moment of panic.
A cross-exchange arbitrage scanneris rolling out next, comparing live prices across Binance, OKX, and Bybit to surface spreads worth a second look — again, surfaced for a human to act on, not executed on its own.
04Built like an exchange, priced like a sandbox
Every position on ZAINEX today is paper trading— real market data and real exchange mechanics (fees, margin, liquidation math) running against a simulated balance, with no real capital at risk. It’s a deliberate choice: a place to pressure-test a strategy, or a model’s opinion of one, before either ever touches a live account.
- StrategiesTiered “Guarantrade” paper-yield strategies — Free, VIP 1, VIP 2, VIP 3 — each locking a chosen allocation at a fixed or variable daily rate, gated by AI credits rather than a paywall.
- WalletA 1:1 wallet-to-AI-credits converter and a three-level referral network, so credits can come from testing the platform or from bringing other traders into it.
- StackA Next.js frontend talking to a Laravel API, with every account, order, and fill persisted to Postgres rather than held in memory — the same durability guarantees a production exchange would demand of itself.
None of this reads like a platform trying to be everything at once. It reads like one still deciding how far “AI-assisted” should go — and, so far, drawing that line on the side of showing its work rather than hiding behind it.