Agentic intelligence for your trading operations.

An entire trading desk, powered by AI — scalable cloud compute and optimisation algorithms searching, validating and deploying strategies at a scale no human desk could match.

Apply for early access Explore capabilities

Trading VPS
Native Python Strategies
Frontier Intelligence

(6) Specialists

Agent Alpha
4 helper runs· temporary
Macro & Rates
Technical Research
Cross-Asset Research
Quant Research
Portfolio Risk

Agent AlphaTechnicalPortfolio Risk

Agent Alpha

Ready for the week of 5 Oct.

Demo account: $25,000
Open positions: 1
Risk in use: 0.3%
Watchlist: 5 markets

Prepare the desk for the trading week. Compare FX and metals, review event risk and our exposure, and use specialists on the strongest candidates. Bring me the analysis and any trade proposal for approval.

I'll read the book, this week's events and the market first, then divide the research.

Running toolsRan 6 tools · 3.9sbalance, 1 position, 14 events, 5 markets, 4 notes

get_situational_awareness→ desk state812ms
get_live_account_stateaccount=demo→ balance 25,000.00live412ms
list_live_positionsaccount=demo→ 1 open · EURUSDlive388ms
econ_calendar_taildays=5→ 14 events640ms
market_contextsymbols=XAUUSD,XAGUSD,EURUSD,GBPUSD,USDJPY→ 5 markets1906ms
agent_memory_searchquery="fx metals weekly"→ 4 notesmemory2410ms
Desk mandateFX & metals · ≤0.5% per tradeGold research · Sephigher-low entriesUSD ↔ goldreal-yield linkEURUSD range notes1.1620–1.1750This week's analysis

The book is light: one EURUSD position, 0.3% at risk. Metals lead the week and EURUSD is boxed in its range. I'll brief four helpers in parallel.

Market overviewH4 · Fri 2 Oct close · illustrative

`XAUUSD`	`4,141.6`	`+0.8%`	Higher low held
`XAGUSD`	`48.62`	`+0.6%`	Tracking gold
`EURUSD`	`1.1684`	`−0.2%`	Range 1.1620–1.1750
`GBPUSD`	`1.3436`	`−0.4%`	Drifting lower
`USDJPY`	`147.85`	`+0.1%`	Range

4 helper runs returnedmacro, events, cross-asset, structure

Gold is the strongest candidate: buyers defended 4,113–4,126 on the pullback and real yields are easing. Silver follows without a signal of its own, and EURUSD has no edge inside its range. Wednesday's FOMC minutes are the main risk, so I've kept the size small.

Technical ResearchXAUUSD H4 · structure and levels1.4s

Price actionXAUUSD · H4 · illustrative

Swing low 4,071Swing lowHigh 4,196High 4,196Higher low 4,113

Entry area 4,113–4,126Entry area

Invalidation · stop 4,099Stop

Target 4,180Target

4,0804,1204,1604,2004,2404,141.64,180.04,122.04,099.0

23 Sep25 Sep29 Sep1 Oct

IdleWorking…Briefing helpers…Waiting for 4 helper runsPreparing proposal…Awaiting your approval.Tasks 5/5 · Helpers 4/4Stop

Tasks

Read the book, events and market context
Brief helpers: macro, events, cross-asset, structure
Rank the candidates, with a reason for each
Chart and levels for the lead candidate
Prepare a proposal for approval

Helpers

Macro & ratesworkingreturned
Macro and rates backdrop for 5–9 OctReal yields easing; USD softer into Wednesday.
Event calendarworkingreturned
Rank this week's scheduled events by impactFOMC minutes on Wednesday is the main event risk.
Cross-assetworkingreturned
Relative strength across FX and metalsGold leads metals; EURUSD and USDJPY rangebound.
Technical structureworkingreturned
Structure across the candidate marketsXAUUSD higher low held at 4,113–4,126.

Trade proposal

XAUUSDBuy limit

Entry: 4,122.0
Stop: 4,099.0
Target: 4,180.0
Size: 0.04 lots
Risk: $92 · 0.37%
Reward : risk: 2.5

Retest of a defended higher low while real yields ease. Size reduced ahead of Wednesday's FOMC minutes.

Risk review · Portfolio Risk

Within account limits
0.67% at risk after fill
Event window: FOMC minutes, Wed

Slide to approve

Message Agent AlphaPrepare the desk for the trading week. Compare FX and metals, review event risk and our exposure, and use specialists on the strongest candidates. Bring me the analysis and any trade proposal for approval.

/Claude Opus 5.5

Anthropic

Claude Fable 5.1

Claude Opus 5.5

Claude Sonnet 5.5

OpenAI

GPT-6 Astra

GPT-6.1 Sol

DeepSeek

DeepSeek V4.1 Flash

DeepSeek V4 Pro

Illustrative workflow · timing condensed

Quant research & cloud compute

Put serious compute behind your research.

Work with your agents in Jupyter. Run CMA-ES parameter searches, evaluate strategies across rolling out-of-sample windows and take approved strategies into cloud deployment.

Jupyter researchExecutable Python notebooks, worked on with your agents.
CMA-ES optimisationAdaptive parameter searches run as cloud compute jobs.
Rolling walk-forward evaluationTrain, then test on periods the search never saw — window after window.

walk_forward.ipynbwith Agent AlphaPython 3 · idle

Session breakout — rolling walk-forward

Fit on 12 months, test on the next 3 unseen months, roll forward each quarter. CMA-ES searches three parameters inside every training window.

wf = experiments.walkforward(    strategy="session_breakout@v4",    symbol="EURUSD", timeframe="H1",    window=("2023-01", "2025-06"),    train="12M", test="3M", step="3M",) wf.search(    method="cma-es",    params={"lookback": (8, 48),            "stop_atr": (0.8, 3.0),            "target_r": (1.0, 4.0)},    population=12, generations=20,    objective="sharpe",) wf.oos_summary()

Out 6 windows · 1,440 evaluations · OOS = unseen test period

Window	Train	Test	IS Sharpe	OOS Sharpe
W1	Jan – Dec 2023	Q1 2024	1.84	1.21
W2	Apr 2023 – Mar 2024	Q2 2024	1.77	0.94
W3	Jul 2023 – Jun 2024	Q3 2024	1.92	1.38
W4	Oct 2023 – Sep 2024	Q4 2024	1.69	−0.22
W5	Jan – Dec 2024	Q1 2025	1.81	1.07
W6	Apr 2024 – Mar 2025	Q2 2025	1.74	0.88

Mean OOS Sharpe 0.88Walk-forward efficiency 0.49

Rolling walk-forward

EURUSD · H1 · Jan 2023 – Jun 2025

6 / 6 complete

W1Train Jan – Dec 2023, test Q1 2024, out-of-sample Sharpe1.21
W2Train Apr 2023 – Mar 2024, test Q2 2024, out-of-sample Sharpe0.94
W3Train Jul 2023 – Jun 2024, test Q3 2024, out-of-sample Sharpe1.38
W4Train Oct 2023 – Sep 2024, test Q4 2024, out-of-sample Sharpe−0.22
W5Train Jan – Dec 2024, test Q1 2025, out-of-sample Sharpe1.07
W6Train Apr 2024 – Mar 2025, test Q2 2025, out-of-sample Sharpe0.88

Train · 12 moUnseen test · 3 moOOS Sharpe

CMA-ES parameter search · W6

240 evaluations · σ 0.30 → 0.04

BestPopulation mean

lookback: 22 bars8 – 48
stop_atr: 1.6 × ATR0.8 – 3.0
target_r: 2.4 R1.0 – 4.0

From notebook to hosted strategy

Research compute runs the experiments. Strategy hosting runs only what you approve.

Research compute
1,440 cloud backtests
Complete
CMA-ES across 6 rolling walk-forward windows
W6 parameters · approved by you
Strategy hosting
session_breakout v4
Running
MT5 · cloud-hosted
Max daily loss
2.0%
Risk per trade
0.5%

Illustrative data Simulated example values for a sample strategy. Not a record of real trading performance and not a guarantee of future results.