Your bot can be wrong.
Your limits can’t.
Your bot has big ideas. Give it hard limits. Guard sizes orders from the stop, caps exposure and halts new entries when your loss limits hit. On your machine, between your bot and Hyperliquid.
Pro: up to 3 accounts. No per-order Guard fee. Compare plans
Confidence: unlimited. Risk: limited.
- RequestedBTC · 1% risk
- Guard’s decisionAllowed
- WhyStop set. Within the limits.
Scripted demo. No orders sent. Illustrative requests under 5× leverage and 2% risk limits. The replay below uses real trades and Guard’s engine.
02 / THE MARKET DOESN’T SLEEP
The market, judged by your rules.
Every dot is a real trade on Hyperliquid, as it happens. For the largest ones, Guard reads the trader’s public account and asks one question: would your limits have let it through?
Why Guard said no
- market not on your list0
- no protective stop0
- leverage above your cap0
- too close to liquidation0
- open risk above your limit0
- trader past the daily loss stop0
- trader past the drawdown halt0
- no stop: Guard set one, sized to your risk0
- resized: risk or size above your limit0
Refused, by market
- BTC—
- ETH—
- SOL—
- HYPE—
- ALTS—
Inspect individual tradesEvery decision, with its reason
Decisions
Open a trade to hold its details while the market continues.
- Time (UTC)
- Trader
- Leverage
- Stop present
- Liquidation distance
- Time (UTC)
- Trader
- Leverage
- Stop present
- Liquidation distance
- Time (UTC)
- Trader
- Leverage
- Stop present
- Liquidation distance
- Time (UTC)
- Trader
- Leverage
- Stop present
- Liquidation distance
- Time (UTC)
- Trader
- Leverage
- Stop present
- Liquidation distance
- Time (UTC)
- Trader
- Leverage
- Stop present
- Liquidation distance
- Time (UTC)
- Trader
- Leverage
- Stop present
- Liquidation distance
- Time (UTC)
- Trader
- Leverage
- Stop present
- Liquidation distance
- Time (UTC)
- Trader
- Leverage
- Stop present
- Liquidation distance
- Time (UTC)
- Trader
- Leverage
- Stop present
- Liquidation distance
- Time (UTC)
- Trader
- Leverage
- Stop present
- Liquidation distance
- Time (UTC)
- Trader
- Leverage
- Stop present
- Liquidation distance
| TIME | TRADER | TRADE | NOTIONAL | LEV. | STOP | LIQ. | HEAT | GUARD |
|---|---|---|---|---|---|---|---|---|
| Waiting for the first trades from Hyperliquid… | ||||||||
Traders are anonymised in your browser. Every trade in BTC, ETH, SOL, HYPE, XRP, DOGE, SUI and AVAX streams from Hyperliquid’s public data; the largest are judged on this page, as many as the public API’s request budget allows.
BLINK. YOU MISSED IT.
Hard limits.
About 3 ms.
That is the median time Guard added per order in our paper test. Your account is already in hand. The decision is ready before your bot can explain itself.
Median bot request to Guard answer, on local SSD in Tokyo, against Hyperliquid testnet. The exchange trip is separate; this is a paper-mode measurement.
Read the measurement →- 0.24 ms
Guard’s own work: check, judge, sign
- 3 ms
Guard adds per order, median¹
- 5 ms
until your order reaches Hyperliquid, from Tokyo
How we measured it
¹ Median of the time from your bot’s request reaching Guard to Guard’s answer, with your account judged from the live account stream and Guard’s decision journal on local SSD; measured 7 Oct 2026 in Tokyo (AWS ap-northeast-1), in paper mode against Hyperliquid testnet. The trip to Hyperliquid is not in it: from Tokyo, 4.5 ms there and back (median), so the order reaches the exchange at about 5 ms.
03 / REWIND REAL TRADES
Big decisions.
Play them back.
You’ve seen the market move. Now rewind one account. Follow a public vault’s actual history, replayed behind Guard’s defaults.
- Requested
- Select a decision above
- Would forward
- Read-only replay
- Reason
- The engine’s decision, under the selected rules.
Playback reveals the recorded curves; it does not rerun the engine. Equity at the cursor is interpolated for display. Summary statistics always cover the full result.
Historical what-if, not a promise of returns. Stops can be missed between recorded prices; later trades stay as traded. Equity at the cursor is interpolated for display.
Replay assumptions and source
A recorded public sample, not a live account or a claim about your bot. Refresh this sample in the full replay.
- Starting equity: the account value Hyperliquid's portfolio history shows at 2026-07-15T23:16:00.027Z, its last point at or before the window start, 47.6 h before it; PnL, deposits and withdrawals in between are not accounted for.
- A what-if: the account's later decisions are kept as they were, although skipping or shrinking a trade could have changed them.
- Inside the window equity is realised: starting equity plus closed PnL, minus fees, plus funding. Open positions are not marked to market there, so a loss inside a trade shows when it closes, and the daily loss stop and the drawdown halt see realised equity only. At the window's end both curves mark their open positions to market at the same price (today's mark when the window ends now; else the latest price seen, a fill or an hourly funding payment's), without costs on either side, so a position the account still holds counts on its curve as Guard's closes count on the guarded one. The realised-only final equity of each curve is given apart.
- Both curves start from the same baseline, marked to market at the window's start: the starting equity is the account value then, and a position held from before the window is measured from its price at the window's start, never from its historic entry, on both curves. The actual curve takes the recorded closed PnL less what such a position made before the window (reported apart per fill, with the as-recorded final equity for reference); the guarded curve holds the same position from the same price until Guard first acts on it. A starting equity backed out of today's value gets that part added back.
- Deposits, withdrawals and transfers move both curves at their time: a deposit by the same amount, a withdrawal by the same share of each account's equity as the replay counts it (realised). With open gains or losses that share differs from the account's real one; the replay follows a withdrawal only while the two stay within a factor of 2 (all withdrawals together), which can still leave the guarded account's absolute equity up to that factor off; returns are time-weighted and unaffected. Returns, drawdowns and the worst day leave them out (time-weighted). The halts see a deposit as live Guard does, which reads the account value: it raises the equity and the peak the drawdown halt measures from, and can hide a day's loss. A withdrawal leaves the halts where they were; live Guard would count it as a loss and could halt on it, which the replay does not credit to Guard.
- Guarded positions exit when and where the account exited, in proportion. A loss is capped at a stop the account placed only through the account's own exits; no stop of the account's is simulated.
- Guarded fees are the recorded fees scaled to the guarded size; a close Guard makes on its own (a fired stop, a flatten, a close at the window's end) is a taker order: it pays the higher of the fee rate of the account's last fill in that coin and the highest rate of its taker fills so far (4.5 bps, Guard's default, before it has shown one), never a rebate, on the value at the higher of its exit price and the price that set it off. Every order of the guarded account is Guard's, so every one of them, the entries and exits it follows as well as its own closes, also pays Guard's 0.02% builder fee; the actual curve pays none. Guarded funding is the recorded funding scaled to the guarded position.
- After a halt, guarded positions close at the next price seen in that coin (the account's next fill, or an hourly funding payment's price), or at the window's end at the window-end price, with half the round-trip cost set for the rules (at least 1 bp) of slippage against them; a fired stop takes its own 1%. Book depth is unknown, so a large close in a thin book pays no more than a small one, which flatters the guarded curve for large closes.
- Leverage and liquidation prices of past positions are unknown: an add is taken to suit the position's isolated leverage, so Guard's check of it (its leverage and liquidation, step 8) is not judged.
- A settlement (a delisted perp Hyperliquid closed at its settlement price, which may be 0) closes the account's whole position there, its loss booked; the guarded position in that coin settles at the same price, with no stop firing, no slippage and no builder fee, and at the worse for Guard of the reported price and the one the account's entry and recorded closed PnL imply (a short reported settled at 0 with a recorded loss settles at the price that loss implies). A short held from before the window and settled at 0 cannot be marked and stops the replay.
- Other positions are valued at the last fill price seen in their coin; a position held from before the window, until its coin trades, at its price at the window's start.
- Guard is taken to run from the window's start and to protect what it finds, as its background sync does: a guarded position that no market stop covers in full (the account's own resting stops and Guard's together) gets Guard's own stop for the whole position, at the default stop distance: at the window's start from the window-start price for a position held from before the window, and later from the latest price seen (before the account's next fill in the coin, or when another entry is judged). A stop of the account's it cancelled while it was needed stays, as Guard's: live Guard refuses a cancel that would leave a position uncovered or push open risk over its budget (judged before the account's next fill in the coin). A stop the account places within 60 s after counts already, as for the judge; the sync looks at most every 5 s. Only market stops protect (a stop-limit can trigger and rest unfilled). A position holding a quantity from before the order history begins (held from before such a window, or an entry not judged for that reason) is left alone: its stops are not known.
- Every entry is judged by Guard's own judge. A trigger order of the account's rests for it if it rested at the entry or was placed within 60 s after it, and protects the position held before; it is the entry's own stop, which Guard sizes the entry from (and refuses the entry for when it is not beyond the price), only when the order history lists it among the entry order's TP/SL children, at the same trigger. Guard attaches its own stop to every other entry, and under "refuse" refuses it. An entry is taken to be sent with its limit at its fill price: a market order from the SDKs sends one about 5% through, which Guard pulls in to 0.5% beyond the mid and sizes from, up to 0.5% of the price more risk a unit. Where the per-trade budget binds, the guarded size is so larger than live Guard's, both ways (bigger gains and bigger losses): 24% with a 2% stop, more with tighter stops (82% with a 0.5% stop). An order that turns a position around is followed as the account's exit and a fresh entry judged from flat; live Guard refuses it as one order, and a bot behind Guard would close and then enter.
- Stop policy "attach": an entry without its own stop is sized from a stop 2% away (rounded to the venue's price grid towards the price), as Guard attaches it, and gets that stop for the size Guard took, never more. Every stop of Guard's (attached to an entry, or placed to protect a position) fires only at prices the replay sees, the account's own fills in that coin, the hourly prices its funding payments imply, and the window-end price: it then closes its own size, no more than is left, at the stop with 1% slippage against it (the replay's assumption for a market stop's fill: a long's stop less 1%, a short's plus 1%; Guard lets its stops fill up to 10% beyond the trigger), tightest first. A partial exit leaves Guard's stops as they were. A dip through a stop that recovers between two of those prices (at most an hour apart while the account holds the coin) is not seen and not stopped, which flatters the guarded curve. A fired stop is taken to fill there even when the account's fill is far beyond it, which flatters gaps and liquidation cascades.
Your bot deserves a circuit breaker.
Go Pro below, or set up pay per order. Every Guard rule stays on.
Show me my own bot’s historyPublic address. No wallet connection.
+47.4% in the 90 days to 15 Oct 2026, through a 2.9% drawdown. Guard takes 0 of its 83 entries as they were, resizes 79 and refuses 4: +23.7%, 8.0% drawdown.
·
- runs entirely in your browser
- no wallet connection · no signature · no keys
- nothing to approve, nothing that can move funds
- we never see the address
Sample: a public Hyperliquid vault · its 90 days to 15 Oct 2026, 22:52 UTC, replayed when this page was built · 587 fills
Guard took nearly all of this account’s entry size: less return, a larger drawdown.
Refused: 4 open-risk cap. Resized: 78 stop added, 1 stop added and sized down.
With deposits or withdrawals in the window, return, worst drawdown and worst day are time-weighted: the flows cut the curve into stretches, and each stretch counts by its own change, so money moved in or out is never counted as gain or loss. A deposit moves both curves by its amount; a withdrawal takes the same share of each.
Equity, 90 days
Account equity · USDC · includes cash flows
- Requested
- Select a decision above
- Would forward
- Read-only replay
- Reason
- The engine’s decision, under the selected rules.
Playback reveals the recorded curves; it does not rerun the engine. Equity at the cursor is interpolated for display. Summary statistics always cover the full result.
No liquidations in this history.
Adjust your replay rules
Every rule applies here, including the account stops.
Difference = Behind Guard − As traded, using the displayed values. pp = percentage points.
The trades that mattered
SELL BTC · Guard’s stop10 Sep 13hstop hit
- As traded
- —
- Behind Guard
- −$1.9k
- Leverage
- —
- Stop
- added
SELL BTC · Guard’s stop08 Oct 16hstop hit (hourly price)
- As traded
- —
- Behind Guard
- −$760
- Leverage
- —
- Stop
- added
SELL ETH · Guard’s stop16 Sep 02hstop hit
- As traded
- —
- Behind Guard
- −$744
- Leverage
- —
- Stop
- added
SELL PUMP · Guard’s stop07 Oct 01hstop hit (hourly price)
- As traded
- —
- Behind Guard
- −$586
- Leverage
- —
- Stop
- added
SELL BTC · Guard’s stop07 Oct 02hstop hit
- As traded
- —
- Behind Guard
- −$417
- Leverage
- —
- Stop
- added
SELL ETH · Guard’s stop12 Oct 17hstop hit (hourly price)
- As traded
- —
- Behind Guard
- −$390
- Leverage
- —
- Stop
- added
SELL SOL · Guard’s stop01 Aug 07hstop hit
- As traded
- —
- Behind Guard
- −$342
- Leverage
- —
- Stop
- added
SELL ZEC · Guard’s stop12 Oct 16hstop hit (hourly price)
- As traded
- —
- Behind Guard
- −$338
- Leverage
- —
- Stop
- added
BTC · open at the end15 Oct 22hmarked at window end
- As traded
- −$801
- Behind Guard
- +$0
- Leverage
- —
- Stop
- —
ZEC · open at the end15 Oct 22hmarked at window end
- As traded
- −$543
- Behind Guard
- +$0
- Leverage
- —
- Stop
- —
ETH · open at the end15 Oct 22hmarked at window end
- As traded
- +$661
- Behind Guard
- +$661
- Leverage
- —
- Stop
- —
| WHEN | TRADE | LEV. | STOP | RESULT · AS TRADED → BEHIND GUARD | GUARD |
|---|---|---|---|---|---|
| 10 Sep 13h | SELL BTC · Guard’s stop | — | added | — → −$1.9k | stop hit |
| 08 Oct 16h | SELL BTC · Guard’s stop | — | added | — → −$760 | stop hit (hourly price) |
| 16 Sep 02h | SELL ETH · Guard’s stop | — | added | — → −$744 | stop hit |
| 07 Oct 01h | SELL PUMP · Guard’s stop | — | added | — → −$586 | stop hit (hourly price) |
| 07 Oct 02h | SELL BTC · Guard’s stop | — | added | — → −$417 | stop hit |
| 12 Oct 17h | SELL ETH · Guard’s stop | — | added | — → −$390 | stop hit (hourly price) |
| 01 Aug 07h | SELL SOL · Guard’s stop | — | added | — → −$342 | stop hit |
| 12 Oct 16h | SELL ZEC · Guard’s stop | — | added | — → −$338 | stop hit (hourly price) |
| 15 Oct 22h | BTC · open at the end | — | — | −$801 → +$0 | marked at window end |
| 15 Oct 22h | ZEC · open at the end | — | — | −$543 → +$0 | marked at window end |
| 15 Oct 22h | ETH · open at the end | — | — | +$661 → +$661 | marked at window end |
How honest is this? A replay is a what-if, not a promise. Skipping or shrinking a trade can change what your bot would have done next, and prices, fees and funding after a skipped trade are taken from your real history. Losses are capped at your stop only where your bot actually had one. Results include fees and funding as Hyperliquid recorded them.
Ready to put Guard in front of your bot?
Choose my Guard ↓04 / PUT GUARD IN FRONT
More thunder.
Less fee.
One flat licence. Zero Guard builder fee. All the rules, however busy your bot gets.
- Zero Guard builder fee
- Every rule. Every limit.
- Your keys stay on your machine
The licence removes Guard’s fee. Venue fees, funding and slippage still apply.
Pro
€149/ month
Up to 3 Hyperliquid accounts
One flat price. No Guard fee per order.
Continue with ProBusiness licences · pay in USDC · VAT may apply. Guard 1.0 is pre-release; your term starts when we send the key.
Prefer pay per order? The default is 0.02% with no subscription. Paper and testnet are free. Compare against your volume → Start with pay per order →
All plans and details →Break-even: Pro pays for itself above about $870k of orders a month, Fund above about $4M. Dollar figures at $1.17 per euro, the rate the prices were set at.
EVERY ORDER GETS THE SAME TREATMENT
Big ideas meet hard limits.
Read the account
Equity, open positions and their stops, from the venue and from Guard’s own record. The riskier view wins.
Size from the stop
Your bot names a stop. Guard turns your risk limit, fees and the venue’s rounding into a quantity.
Check every limit
Leverage, open risk, the daily loss stop, the drawdown halt. One fails, the order never leaves.
Send, stop, record
The order goes out with its stop resting on the venue, and the decision is written to a checksum-chained journal that shows if a line was changed or removed.
“Trust me, bro” is not a risk policy.
Risk at the stop sets the size. Leverage and liquidation distance limit the exposure. Daily loss and drawdown limits can halt new entries.
Read all nine rules and their defaults
The nine rules
Four refuse an order, two halt new entries, two shrink it, and one adds the stop your bot forgot. You set each one; the defaults are Guard’s policy.
- Market allowlistdefault all markets
- REFUSES · Only markets on your list can be traded.
- Required stopdefault Guard sets one
- ADDS A STOP · No stop? Guard sets one, sized to your max loss at the stop. (Or refuses, if you choose.)
- Max leveragedefault 5×
- REFUSES · Refuses an order that would take the position above your leverage cap.
- Min distance to liquidationdefault 10%
- REFUSES · Refuses an order whose position would sit closer than this to its liquidation price.
- Max open riskdefault 6%
- REFUSES · Refuses a new trade when the loss at all stops, open positions plus this one, would pass this share of the account.
- Daily loss stopdefault 6%
- HALTS · Once the account is down this much since the start of the UTC day, new entries stop until the day is over.
- Drawdown haltdefault 25%
- HALTS · Once the account is this far below its peak, new entries stop until a human resumes.
- Max positiondefault 200%
- RESIZES · Shrinks an order whose position would be larger than this share of the account.
- Max loss at the stopdefault 2%
- RESIZES · Shrinks the order until the loss at its stop, fees included, is at most this share of the account.
Always, whatever you set: stops only tighten, a restart never resets a stop, and a drawdown halt waits for a human.
RESEARCH · DATA TO 6 OCTOBER 2026
Hard limits.
Smaller drawdowns.
The defaults cut the median account’s max drawdown on the held-out replay. Same trades, judged behind Guard’s rules. Here is what changed.
- −43 pts
max drawdown, median account
cut by 43 points (95% interval 23–58)
63 accounts held out
- −67% → −17%
worst day, median account
as traded, then behind the defaults; no interval computed for this difference
63 accounts held out
- 60%
return per unit of drawdown improved
share of accounts (95% interval 48–73%)
63 accounts held out
Original test split: 63 accounts, corrected after the original results had been seen. Rules and membership stayed fixed; this is not a fresh blind test. Medians; 95% bootstrap intervals. A mixed sample: ordinary active traders, long-term winners and recently liquidated accounts. Post-hoc filter sensitivity: excluding 2 now-ineligible test accounts leaves 61 accounts and 61 of 171 events flat (35.7%; 95% interval 15.5–54.1%). Those excluded accounts supply 62 of the primary split’s 123 flat events. We tested 5,940 registered rule sets on the real trades of 262 Hyperliquid accounts. The corrected replay did not support changing our defaults.
The numbers come with receipts.
Replays are not a promise of returns. The best traders gave up some gains behind the defaults; stops touched between fills flatter the replay. The original defaults sample is mostly manual traders; a separate cohort now covers 148 likely bots.
Read the trade-offs and replay limits
It costs the best traders part of their gains: in the replay the median long-term winner’s return fell from 54% to 48%. Opening the halt alone does not recover that median return. With all five engine limits opened, the median is 44%; attached stops, the position cap and costs still apply. This does not isolate their separate effects. 9 winners, training and validation accounts.
In the original 63-account test split, the median account’s full-size entry pass-through was 19% (95% interval 2–32%). Separately, across 76,071 pooled entries from 199 training and validation accounts, 39% were held by a halt (35% by the drawdown halt, once an account was already 25% down), 21% refused and 11% resized.
What-if replays: each account’s own trades over the 180 days to 6 October 2026, judged by Guard’s engine; later decisions are kept as traded.
These replay studies cover main-dex perpetuals; HIP-3, spot and outcome markets are excluded. The liquidation autopsy covers 4 Sep – 3 October 2026, read at a daily snapshot.
A replay of each account’s last 180 days to 6 October 2026 through Guard’s engine. “Beat” means our pre-registered score, with its constraints, on accounts the rule set was not chosen on; the best challenger scored +0.092 (95% interval −0.15 to +0.29). The original test split was rerun after unblinding to correct historical-data defects; this is not a fresh independent confirmation.
Kick the tyres. Bring your bot.
Start with a public address. No wallet connection or installation needed for the browser tools.
Guard
Source availableRisk firewall between any bot or AI agent and Hyperliquid.
Backtest
Try todayReplay any address under your rules.
Watch
Try todayYour bot’s trades judged live in the browser.
Live market
Try todayHyperliquid’s trades judged by your rules, anonymised.
What else we are building and exploring
- Agent kit (MCP)At first release
Guarded trading tools for Claude, Cursor and other MCP clients.
- MonitorPlanned
Pair the browser with your local Guard: decisions and kill switch.
- DataPlanned
Point-in-time Hyperliquid data: trades, books, node gossip.
recorded since Oct 2026
- LabsPlanned
Research notes: honest backtesting, the signal lab.
- Guarded ArenaPlanned
AI agents trading in public, every veto visible.
- MintExploring
Toolkit for HIP-3 market deployers.
- TerminalExploring
Guard in the browser for manual traders.
Live: works today. Building: in progress. With Guard 1.0: ships with Guard. Planned: decided, not started. Exploring: an idea we are testing.
Good questions. No smoke machine.
Can I try it without connecting a wallet?
Yes. Replay and Watch use a public Hyperliquid address in your browser. They do not place orders.
Can Guard guarantee I will not lose money?
No. Limits constrain exposure; fees, funding, slippage and price gaps still matter. The studies explain the observed benefits and their limits.
When can I install Guard?
Read the public source and prepare your setup. Signed installers are next; Replay and Watch work today.
Put a firewall in front of your bot.
Guard’s source is public. Join for signed release updates. The browser tools are available to try today.
0.02% of each order’s value that Guard sends to Hyperliquid · no subscription · or a flat licence. Pricing
Launch offer: join before Guard 1.0.0 and get Pro free for 3 months.