The machine that has to prove it can winbefore it is allowed to trade a single real dollar.
An AI-run forex lab on one laptop with one GPU. It searches hundreds of millions of trading strategies, runs a team of AI analysts, forward-tests the best ideas on live prices, and keeps real money locked behind hard safety rails until the evidence is real.
The control room, as the owner sees it.
The same Trading Desk dashboard that runs on the lab's PC, mirrored here view-only: which agents ran and when, the demo bot's fills, and how many trades the safety gate checked and passed today.
Demo account: recent orders (practice money)
Live activity
AI research team
Grow $25 into $250. But only with an edge that survives the truth test.
Most trading bots look brilliant on the data they were tuned on. This lab is built the other way round: every idea is judged on data it has never seen, then on live prices it could not have known. If nothing passes, nothing trades. So far, honestly, nothing has passed, and that is the system working.
Bot capital
Status of real money
What would unlock it
A GPU that never sleeps, hunting for an edge.
An RTX 5060 tests rule-based strategies and small neural networks across years of hourly and 15-minute prices. Survivors must win across separate time periods, stay profitable when their settings are nudged, and finally pass one exam on currency pairs the search never saw.
Round 1 · 48-hour search
Round 2 · 5-fold + unseen exam
Round 2 progress
Strategy families on the leaderboard
Current leaderboard · top strategies by robustness score
R = one unit of risk (the planned loss if the stop is hit). +0.10R per trade on average is a real edge; the five bars show the result in each separate time period, so a good strategy should be green in most of them.
100 strategies, real live prices, fake money.
The best ideas from the research trade in a simulator against live market prices every few minutes, across 17 currency pairs. Nothing here is back-filled or edited. A strategy is judged only at fixed checkpoints (40, 80, 120, 200, 300 trades) by a statistical test fixed in advance that corrects for luck across every strategy being tested, and it must also earn at least +0.10R a trade, spread across several pairs.
Cumulative result of every forward-test trade (R)
Result by currency pair
Candidates closest to graduating
Live trade feed
A room full of AI analysts, arguing about every currency.
Separate AI advisors read the news, central-bank calendars and price behaviour; a risk manager checks exposure and volatility; a watcher reads prices every second and can pause everything on a spike or a news release. Their views are logged and graded against what the market actually did.
Desk risk
Geopolitical risk
Market volatility
Council verdicts per pair
Currency strength · news sentiment vs central-bank policy
Coming up · economic calendar
Latest AI market brief
Watcher alerts
The real account stays private.
Detailed real-account figures are no longer published here. The bot's real-money status and its count of real trades are shown in the Mission section above. Everything else on this page is research, simulated or practice data.
Practice (demo) bot
Every agent, live.
The lab is a crew of small programs and AI agents, each with one job, supervised by a dashboard that restarts anything that falls over.