Deploy a Hyperliquid perps bot that trades your strategy — on the liquidity retail can't see.

Most "whale tracking" tools watch a handful of hand-picked wallets. We don't. A dedicated data pipeline pulls the entire active Hyperliquid leaderboard — roughly 21,000 accounts — and re-sweeps it about every 10 minutes, reading each account's positions and resting orders directly from Hyperliquid.

From that raw data it builds four anonymized layers: forced-liquidation clusters (priced from each position's actual liquidation level), declared stop-loss pools and take-profit walls (classified by real order type, not guessed from price), and aggregate whale positioning — net open-interest skew, a whale-versus-retail split, fresh inflows, the share of positions trapped underwater, and crowd leverage. Market-maker grid bots are detected and excluded, because their hedged inventory isn't directional and would otherwise smear the signal.

The output is a heat signature on the price tape: where the fuel is, where the stops are pooled, and which way the real size is leaning. It's the same order-flow picture professional desks pay for — and it's public on /watch.

An AI that's sure of itself is not the same as a safe one. So the safety lives outside the model, in code the model can't reach.

Every entry is placed with its stop-loss in a single signed order — so there is never a moment where a position exists without a stop. The two take-profits attach immediately after as standalone reduce-only orders, kept deliberately out of the entry-and-stop group so that a take-profit filling can never cancel the stop on the rest of the position. The stop is anchored to structure (the swing or level that invalidates the idea), not to a fixed percentage that noise can wick. Before any order is even constructed, the setup has to pass server-side risk/reward and expectancy checks; a setup that fails them is rejected, regardless of how confident the reasoning sounds. Above the per-trade layer, account-level kill switches stop the whole bot after a daily loss cap or a losing streak — the failure mode where every individual trade looks fine but the account still bleeds. After the first target fills, the stop ratchets to breakeven on the remainder. Position size is derived from your risk-per-trade setting and the stop distance, so a wider stop automatically means a smaller position.

You can hand the bot an aggressive strategy in plain English. You cannot hand it one that removes these rails.

Non-custodial isn't a marketing word for us — it's the architecture. Your funds sit in your Hyperliquid account. The bot operates through Hyperliquid's native agent-wallet system: you grant a dedicated agent key permission to sign trades and manage stops on your behalf, and Hyperliquid's protocol explicitly forbids that key from withdrawing or transferring funds.

That means the worst case is bounded. If our servers were compromised tomorrow, an attacker could place trades — annoying — but could not move a dollar out of your account. You can inspect exactly what the agent is permitted to do, and revoke it instantly, on app.hyperliquid.xyz. No deposit to us, no shared custody, no trust-us promise.

Hyperliquid is the leading on-chain perpetual-futures exchange — a fully non-custodial DEX with deep liquidity, sub-second execution, and no KYC. Perps let you go long or short with leverage; "perpetual" just means there's no expiry. It's where HyperPerps AI trades, because its agent-wallet system is what makes a non-custodial bot possible in the first place.

If you're starting fresh, open an account with our referral for a 4% fee discount, fund it with USDC, and you're ready to deploy. Compare Hyperliquid bots for 2026 →