THE 02 / 03

The trading coreunder every signal.

No SaaS middleman. No broker. Every venue is a Rust adapter compiled into the same hot path. Spot, perpetuals, futures, options, and prediction markets, all on the same nanosecond timestamps.Call the engine through a stable API, write strategies in Rhai or Python, or describe what you want in plain English and let the AI pipeline draft and patch the code.

← Back to home
95ns · dispatch · hot path
p951.2 µsp993 µs
·MELAYA·v2026.06

Trade everything that prints a price.

Connected Venues

Trade across 6 venues

Spot, futures, perpetuals on one execution layer, one ledger, consistent risk controls.

Hover any logo to inspect coverage
Crypto venues
spot · perp · futures · option
Financial markets
FX · indices · equities
Prediction venues
Polymarket · Kalshi · Drift · SX · Azuro · Overtime
Bar timeframes
1m · 5m · 1h · 1d

One strategy file. Every market shape.

125x
MAX LEVERAGE · PERP

The same strategy file fires emit_long on a Binance perp, emit_limit_short on a Coinbase spot, set_sl_price on a Hyperliquid future, and a YES / NO / scalar payload on a Polymarket prediction. Same Rust hot path. Same risk verbs. Different venue, identical trades[] shape.

MARKETemit_long · emit_short
LIMITemit_limit_long · emit_limit_short
STOP LOSSset_sl_price · atr_trail
TAKE PROFITset_tp_price · partial
TRAILatr_trail · pct_trail
PREDICTION MARKETS · OUTCOME
YESNOSCALAR
Order verbs
5 · MARKET to TRAIL
Market shapes
spot · perp · futures · option · prediction

Write it. Or describe it to AI.

Two paths to the same engine. Write the strategy yourself in Rhai or Python with 80+ runtime functions, the same hot path that runs in production. Or describe the strategy in plain English and let the three stage AI pipeline draft the code, critique its risk handling, and patch the gaps. Either way you ship the same .rhai file to the same backtest and the same dry-run mode. Live execution is gated pending licensed execution.

DRAFTER
"BTC perp, 1m. Long when EMA(8) crosses EMA(21) and RSI(14) > 50. 0.5% stop-loss."
fn evaluate() {
let e8 = ema(closes("BTCUSDT", "1m"), 8);
let e21 = ema(closes("BTCUSDT", "1m"), 21);
if crossover(e8, e21) && rsi(14) > 50 {
emit_long(qty);
set_sl_price(price() * 0.995);
}
}
CRITIC
+uses set_sl_price (managed risk)
+no banned tokens
xmissing take-profit, add set_tp_price
xtp_pct not in paramsSchema
// + add take-profit, tunable params
set_tp_price(price() * param("tp_pct"));
set_state("last_entry_bar", bar());
{ }
REPAIRER
ema_rsi_crossover.rhaivalidated · saved
Runtime functions
80+
Languages
Rhai · Python
AI pipeline
3 stages

Replay. Optimize. Re-replay.

1.84
BEST SHARPE · iter #6 of 50

Hit BACKTEST and the same Rust engine that runs production replays your strategy against historical bars. Twelve metrics come back: Sharpe, Sortino, Calmar, max DD, win rate, profit factor. The equity curve plots over OHLCV candles.

When one run is not enough, hand it to the AI Optimizer. Pick an objective. Pick parameters to vary. The optimizer spawns an LLM agent that proposes a set, runs a backtest, reads the result, reasons about why, and proposes the next. Up to fifty iterations per run, agent reasoning visible on every step.

OBJECTIVESHARPE
WINDOW30 D · 1H
ITER6 / 6
EQUITY · 90 D · 1H+28.4%
#PARAMSSHARPERETDD
#6len=42 tp=0.62% sl=0.31%1.84+28.4%−4.9%
#5len=39 tp=0.55% sl=0.28%1.61+24.7%−5.6%
#3len=36 tp=0.50% sl=0.25%1.22+17.3%−7.4%
#1len=24 tp=0.35% sl=0.20%0.71+8.9%−10.4%
WIN RATE61.2%
PROFIT FACTOR2.18
TRADES144
Backtest metrics
12
Optimizer objectives
6
Iterations / run
up to 50

The AI keeps tuning.
You keep the keys.

QUANT
MARKET
AWAITING APPROVALiter #4 · sharpe 1.61
tp_pct0.00400.0062
sl_pct0.00300.0024
cooldown_bars86
Approve & ApplyReject & Halt
TRADER
RISK

A four-persona crew runs beside the live strategy. Quant reads execution metrics. Market classifies the regime. Risk runs a walk-forward backtest on the proposal. Trader composes the new parameters. The status endpoint polls every four seconds. When the AI wants to widen a take-profit, it surfaces the diff and the rationale. APPROVE & APPLY pushes the new params on the next heartbeat. REJECT & HALT stops the optimizer. The default is a human in the loop.

Crew
Quant · Market · Risk · Trader
Status refresh
4 s
HITL default
approve every diff

Monitor every live strategy. Stop any of them in one tap.

The same in-house dashboard you see here runs every live strategy. Positions, fills, equity, drawdown gates, dry-run status, paused orders, one cockpit. Watch a crew reason, watch the engine fire, then pause, adjust, or kill mid-run without leaving the page.

MelayaUSDTMELAYA/USDT63.9k−0.89%MELAYA
100
Citadel connections
20
Streams / connection
20
Authenticated IP cap

Per-tier scope is the safety rail. Keys live in Infisical, never in the database. Private streams (balance, orders, fills, positions) multiplex on one /ws/private socket per venue. One JWT, one socket, control frames in flight. No reconnect storm. No fan-out cost on the operator's box.

Sandbox
Conns
2
Streams
1 / conn
Throttle
500 ms
Depth
L10
Forge
Conns
10
Streams
3 / conn
Throttle
200 ms
Depth
L50
Bastion
Conns
20
Streams
5 / conn
Throttle
100 ms
Depth
L100
Citadel
Conns
100
Streams
20 / conn
Throttle
real-time
Depth
full L2
Key slots / venue
multi
Secrets backend
Infisical
Probed capabilities
stream · feed · execute
"One engine to rule them all. And in the darkness bind them."
Join the community
// Cookies
Melaya uses a small set of first-party cookies that are strictly necessary to authenticate you, maintain your session, and protect the platform from abuse. We do not use advertising cookies, cross-site trackers, or third-party analytics by default. The full cookie list is in our Privacy Policy.