In deep build mode · Pre-launch

Z2H

Private AI Trading Terminal for Institutional FX

// ZERO TO HUNDRED — 0% MANUAL, 100% AI

Talk to founder →
SCROLL ↓
/* THE PROBLEM */

Institutional desks face an impossible
tradeoff today.

Cloud AI promises alpha but leaks data. Legacy EAs are stable but stuck in the 1990s. Most desks are choosing between performance and privacy — and losing both.

01

Cloud AI = Data Leak

Sending order flow to OpenAI or Anthropic means your strategy lives on someone else's server. Compliance teams say no.

02

1990s EAs Still Run Most Desks

Grid logic, martingale, hardcoded SL/TP. One platform freeze and your account is gone. EAs fail challenges constantly.

03

"Black Box" = Regulator's Headache

"The AI decided" is no longer an acceptable answer. SEC, FSC, and EU AI Act demand explainability for every trade.

/* THE BET */

Two layers.
Zero compromise.

Z2H separates safety from intelligence. Layer 1 is deterministic and auditable. Layer 2 is an LLM operating only inside Layer 1's hard guardrails.

// LAYER 01

Physical Safety Floor

  • Kill Switch (instant stop)
  • Daily Loss Floor (-2.5% / -5%)
  • Order Idempotency
  • NaN/Inf Validation
  • Timeout + Restart
  • 100% Deterministic, fully auditable
// LAYER 02

Full LLM Discretion

  • Market Context Analysis
  • Dynamic Strategy Selection
  • Entry / Exit Decisions
  • Memory Learning
  • Feedback Loop
  • Operates only inside Layer 1's bounds
/* WHERE WE ARE */

Built. Validated.
Scaling.

Real numbers from real builds. No hype, no inflated metrics.

⚡

GPT-4 Fine-tune

77%

Accuracy in 4-hour live test window. Validated alpha exists.

📊

Forward-test

62.8%

14-month MT5 win rate, 835 trades, 7/7 segments profitable.

🛡️

Architecture

L1+L2

MT5 connector + Layer 1 safety + Layer 2 LLM live in code.

🚀

Next Milestone

Qwen 7B

Fine-tuning open-source LLM to match GPT-4 baseline for on-prem.

/* HONEST GAP DISCLOSURE */

Backtest ≠ Live. We know.

Our 1.5B model shows a real backtest-to-live performance gap. That's why we're scaling to 7B → 70B parameters with proper SFT — and being upfront about the gap before any conversation. Most vendors hide this. We won't.

Building the next era
of institutional AI trading.

If you're a quant trader, prop desk PM, or institutional FX practitioner — I'd love to learn from your perspective.

founderkim@zero2h.com