Automated execution. Your account. Your rules.
AI Trading Fleet connects to a broker or exchange account you already hold and executes the strategy and risk limits you configure. It never takes custody of your funds, and no strategy runs unrestricted live until it clears a published validation gate.
row nsha256( prior_hash || canonical_json(payload) )row n+1stores prior_hash + its own hash, written atomicallyUTC daySHA-256 Merkle root over every row that dayAlways on.
One control halts every strategy, instantly.
The Fleet never trades a strategy or a limit you haven't set yourself. It runs on your account, inside your limits, and stops the moment you say so.
Four steps. Nothing hidden.
Link your account.
Link a supported broker or exchange account with a trade-only API key. Withdrawal-capable keys are rejected at connection.
Set your own limits.
Pick a strategy, labeled by its real validation stage, and set your own position size, exposure, and loss limits.
Let the rules run.
The Fleet executes inside the limits you set — server-enforced, not a suggestion the frontend can be talked out of.
Stay in control.
Watch fills, exposure, and drawdown on your own dashboard. One button halts everything, instantly.
Join Early Access →No code. Just ask.
Describe the idea in plain words. A team of specialist agents takes it from there: one drafts the rules, one researches, one backtests, one checks risk, one paper-trades. You get backtested strategies in hours, not weeks of coding.
- Plain-English requests, no code
- Agents work together, one per stage
- Backtests with costs included
- Every order still passes your limits
Find me a strategy that fades Polymarket prices after they overreact to breaking news.
Drafted three candidate rules around post-news reversion. Sending the strongest one to research.
Pulled the market history and tested the rule on data it never saw. One candidate dropped for a thin sample.
Walk-forward backtest finished, costs included. The full report is on your dashboard.
Ran every simulated order through propose() against your limits. Nothing breached.
Trading it on live data with simulated fills. It moves to live only after it clears the gate.
Every strategy earns its stage.
A strategy moves forward only when it clears fixed, objective criteria — never on a hunch, never because launch day arrived.
Development
Written, unit-tested, no market data run yet.
Backtest
Walk-forward tested on historical data, costs included.
Paper
Live market data, simulated fills, no real order sent.
Validated
Brier < 0.22, win rate ≥ 58%, Sharpe ≥ 1.5, minimum sample size.
Controlled Live
Real orders, capped size, ≥30-day testnet track record required first.
General Live
Full position sizing within your configured limits.
Six markets. One set of rules.
Every asset class runs through the same risk engine, the same limits and the same audit chain. Pick one to see what the Fleet watches there. The tag on each is what's true today.
- Trade-only keyok
- Per-venue capok
- Daily loss haltarmed
- Kill switchlive
propose()One platform. Many markets — built out honestly.
Every panel below is marked by what's actually true today, not what's planned.
Bybit
Polymarket
Kalshi
Betfair
Global brokers
Roadmap, not a promise for launch day: the plan is to keep adding crypto exchanges, stock/index brokers, and prediction markets globally — any venue that supports a trade-only API, not limited to any single country — and to bring Polymarket and Kalshi up from signal-only to full order placement, same trade-only API-key model as Binance and Bybit, once each integration earns it. Nothing here is live until its panel above shows an "Available" pill.
Every market climbs the same ladder.
Crypto, prediction markets, sports exchanges and brokers all run through one pipeline: signal, paper, testnet, live. A venue moves right only on evidence. This is how far each one goes at launch.
- < 0.22Brier score
- ≥ 58%Win rate
- ≥ 1.5Sharpe ratio
- 90 daysRolling window
- ≥ 30Resolved predictions
All five must hold at once. Passing writes a Graduation_Gate row to the audit chain, and the next stage refuses to start without it. Thin samples never graduate.
Limits you set. Limits the server enforces.
Every item below is a real, configurable field in the Fleet's risk engine — not a marketing promise.
-
Max % of NAV exposed
Hard ceiling across the whole account.
Capital at risk -
Max % per strategy
No single strategy can dominate exposure.
Per strategy -
Max % per venue
Caps concentration in one broker or exchange.
Per venue -
Max size per trade
Set as a % of capital or an absolute cap.
Per position -
Halt after max daily loss %
Trading stops for the day, automatically.
Daily loss -
Halt at max drawdown %
A circuit breaker, not a suggestion.
Drawdown -
Minimum required to trade
Below your threshold, the Fleet stays flat.
Edge / confidence floor -
Stop everything, instantly.
One control, always visible, always live.
Kill switch -
Trade-only API keys
Withdrawal-capable keys are refused at connection.
Key scope
Not marketing copy. Real modules.
A partial look at the actual code enforcing the rules above — dependency-injected, no global state, every trading decision behind a pure function.
Trace an order through the pipeline.
One console, four slices of the real module layout. Pick a view.
3 strategiesEach one carries its real validation stage — Development, Backtest, Paper — and moves forward only when it clears fixed criteria.propose()Pure function — no I/O, no clock, no randomness. Same inputs, same decision, every time.KillSwitchStateful wrapper around propose() — the control behind "kill switch, always on."AuditChain.append_row()sha256(prior_hash + payload), written atomically — a hash chain, not an editable log.verify_isolation_or_exit()Checks cloud project, database, credentials, and inbound link at startup. Any violation: audit, alert, exit(1).load_phase()Refuses to start a phase without a passed Graduation Gate audit row from the phase before it.- prior_hash
- hash of row n−1
- hash
- sha256(prior_hash ‖ canonical_json(payload))
- write
- both columns, one atomic insert
- daily root
- SHA-256 Merkle root per UTC day
- edits
- append-only
Illustrative mockup of the real module layout — not a live account.
Every order passes through isolation checks, then the pure risk function, then the audit chain — in that order, every time.
No promises. Just rules.
Six rules the Fleet is built around. Each one lives in code, and each one below says where.
No custody.
The Fleet never holds your funds. It places and cancels orders on an account you already own.
Trade-only keys.
Withdrawal-capable API keys are refused at connection, so the Fleet has no path to move money out.
Pure risk decisions.
Every order goes through propose(): no I/O, no clock, no randomness. Same inputs, same decision.
Hash-chained audit.
Each audit row stores sha256(prior_hash + payload), written atomically. An edit breaks the chain, so it shows.
Earned stages.
Nothing reaches live trading until it clears a published, fixed validation gate — not a launch date.
Kill switch, always.
One control halts every strategy instantly. Always visible, always live.
Join now, or pay full price later.
The only real advantage here is time. Early access is priced at half of launch pricing for as long as the early-access window runs.
- Reserves your spot on the waitlist today
- Full early-access price (50% of launch price) disclosed after launch
- Access from launch day, 2026-10-23
- Testnet execution on Binance & Bybit at launch
- Everything in Early Access, same 50%-of-launch-price terms
- Plus 5 invite codes, generated the moment your deposit clears
- Each code lets one person you send it to skip the waitlist entirely
- For prop desks, trading groups, or anyone bringing their own team
Early Access is a $10 deposit; the Syndicate Pass is $19 and additionally mints 5 invite codes. Either way, it's what locks in your spot, not the final subscription price — that's disclosed after the 30-day launch.
Reserve your spot
Reservations run through Telegram, not a web form — it's where the invoice and your payment confirmation land anyway. Tap a tier, the bot asks your email (optional, so we can reach you before launch), then hands you the payment link — $10 for Early Access, $19 for a Syndicate Pass.
A reservation, not a charge. The bot shows the exact price and terms before you pay anything. Already have an invite code? Message the bot and send /redeem <code>.
Questions worth asking
before you connect a key.
The things people ask before trusting software with a trading account — answered plainly.
Does AI Trading Fleet ever hold my money?
No. You connect a trade-only API key to your own broker or exchange account. The Fleet places and cancels orders inside limits you set — it never takes custody, and withdrawal-capable keys are rejected at connection.
What actually runs live on launch day?
Binance and Bybit connect in testnet mode at launch. Every other market listed above is signal-only or paper/backtest until its integration and strategy both clear their validation stage — shown honestly in the Markets section above.
Is Early Access a real charge or just a waitlist?
A token deposit, paid in crypto via Telegram, reserves your spot right now — $10 for Early Access, $19 for a Syndicate Pass. It's a deposit toward your eventual price, not the full subscription charge — that full price is disclosed after the 30-day launch.
What happens after the 3-month early-access window?
New signups move to public pricing — the full launch price, no discount. Early-access and Syndicate users keep their locked-in terms for the window they signed up for.
What's a Syndicate Pass, and how do the invite codes work?
$19 — same terms as Early Access, plus 5 unique codes sent to you the moment payment clears. Send a code to anyone — they message the bot with /redeem <code> and skip the waitlist entirely, no deposit needed. Each code works once.
Can a strategy lose money?
Yes. Trading involves real risk of loss. No strategy on this page is marketed as guaranteed, and nothing is promoted past Paper stage without meeting a published, fixed evidence bar.