Hyperliquid vaults vs diversified non-custodial copy trading

Two ways to allocate on Hyperliquid

If you want exposure to someone else's trading on Hyperliquid, you have two realistic structures. You can deposit into a vault, which is a pooled account that a leader trades on behalf of every depositor. Or you can keep your capital in your own account and let a system mirror selected leaders' trades into it under a permission you granted and can revoke.

These are often discussed as if they were the same product with different branding. They are not. They differ on custody, on how positions interact, on how fees accrue, on what you can observe, and on what you have to do when the strategy behind them decays.

This note sets out those differences precisely. It does not argue that one structure wins. The honest answer is that the right structure depends on your capital, your time horizon, and how much operational control you actually want.

How Hyperliquid vaults actually work

A vault is an on-chain account with a leader and a set of depositors. The leader trades the vault's capital. Depositors receive a share of the vault proportional to what they put in, and their share appreciates or depreciates with the vault's equity.

Three properties follow from that structure, and they matter more than any marketing description of a particular vault.

First, capital moves. Depositing means transferring USDC out of your own account and into the vault's account. From that moment you hold a claim on the vault, not a balance you control directly. Withdrawal is a request against the vault's liquidity, subject to whatever lockup or delay the vault enforces.

Second, the vault is one book. Every position the leader takes lives in the vault's single margin account. There is one net position per market by construction, one liquidation price per market, and one shared pool of margin backing all of it.

Third, the leader is normally compensated with a profit share on gains, and leaders are typically required to keep their own capital in the vault alongside depositors. That alignment is real but partial: a profit share pays on the upside and does not claw back on the downside.

• Custody: capital sits in the vault account, not yours.
• Structure: one leader, one strategy, one net book.
• Exit: withdrawal request subject to the vault's lockup and liquidity.
• Compensation: profit share on gains, plus the leader's own stake.

How non-custodial basket copy trading works

The basket structure inverts the capital flow. Your USDC stays in your own Hyperliquid account. You grant a trade-only agent approval — trade-only agent approval; no withdrawal or transfer rights — which lets a system place and manage orders on your behalf without any ability to move funds out.

Instead of one leader, several qualified leaders are mirrored at once. Each runs in one Hyperliquid sub-account per mirrored leader, so a long from one leader and a short from another in the same market coexist rather than cancelling. Capital is allocated by score rather than split evenly, and the whole basket is bounded by 10 leaders at full diversification.

The trade-off that comes with this is capital fragmentation. Each sub-account needs enough margin to carry a meaningful position, which is why full diversification only makes sense above a volume threshold and why smaller accounts run a single sleeve until they cross it.

Custody and control

The difference here is categorical, not a matter of degree. In a vault your funds are in an account whose withdrawal logic you do not control. In the basket structure your funds never leave your address, and the permission you granted cannot withdraw or transfer them.

Control also shows up in how you stop. Leaving a vault is a withdrawal request that clears on the vault's timetable. Stopping a mirrored basket is revoking an approval, which takes effect immediately and leaves your positions where they are for you to manage or close yourself.

Non-custodial does not mean risk-free. Your capital can still be lost to the market, to leverage, and to bad execution. It means the specific class of counterparty and withdrawal risk that comes with pooled custody is absent — nothing more.

Position netting and isolation

A vault holds one net position per market. That is a design feature when a single leader is running a single coherent strategy: it minimises margin usage and keeps the book simple. It becomes a defect the moment you want exposure to more than one strategy at a time, because opposing views collapse into a single line.

Deposit into three vaults and you do get three books, but you also get three separate deposits, three lockups and three withdrawal processes to manage, and no unified view of your allocation.

A basket with per-leader sub-accounts is the other extreme: more accounts to margin, but each leader's exposure, PnL and liquidation distance stay measurable on their own. That measurability is what makes scoring, weighting and replacement possible at all.

Leader selection and what happens when a strategy decays

In a vault you make one selection decision at deposit and then live with it. If the leader's edge decays — a funding-carry strategy after the basis flips, a momentum book in a chopping quarter — the decay arrives in the vault's equity curve first and in your decision-making second. Your remedy is to withdraw, on the vault's timetable.

In a scored basket, decay is supposed to be handled by the system rather than by you. The Elite Top 10 is a sticky basket, and a higher score elsewhere never forces a leader out on its own — removal follows fixed rules instead: emergency triggers such as account value under roughly $1,000 or extended inactivity remove a leader immediately, while soft issues such as thin activity, jump-adjusted drawdown above 35% over 30 days, or ROI below -15% accrue at most one strike per UTC day and require three strike-days before replacement. The replaced sleeve is closed and its capital is redeployed while the other sleeves are untouched.

The honest caveat is that replacement is reactive. No scoring system detects decay before it produces losses; it detects it earlier and acts on it mechanically instead of emotionally. That is an improvement in latency and discipline, not a form of foresight.

Fee models

Vaults typically charge a profit share: a percentage of gains, usually with high-water-mark accounting so the same gains are not charged twice. You pay nothing in a flat month and pay materially in a strong one.

The basket structure here charges 0.1% of mirrored notional volume, applied at execution through Hyperliquid's native builder-fee mechanism, on opens and closes alike. You approve a maximum rate up front and it is revocable with the agent approval.

Neither model is strictly cheaper. A profit share costs nothing when the strategy is flat and a great deal when it is up. A volume fee costs something whenever the system trades, including in a flat or losing period, and scales with turnover rather than outcome. A low-turnover strategy that returns well is cheaper on volume fees; a high-turnover strategy that goes nowhere is cheaper on a profit share.

What matters more than the headline rate is what else you are paying. Exchange taker fees, spread and funding apply in both structures and are usually larger than either management fee.

When a vault makes more sense

There are real cases where the vault is the better structure, and pretending otherwise would be dishonest.

If your capital is small, a vault gives you exposure to a full strategy at a size that would be impossible to replicate across several isolated sub-accounts, where minimum order sizes and margin requirements would round your positions into noise.

If you have high conviction in one specific operator and want their strategy rather than a blend, a vault gives you exactly that with no tracking error: your share moves with the vault's actual fills, not with a mirrored approximation of them.

And if you want zero operational surface — no approvals, no sub-accounts, no rebalancing, no readiness checks — a deposit is simpler than anything a mirroring system can offer.

• Small capital that cannot support several isolated sleeves.
• High conviction in one specific leader rather than a blend.
• A preference for no operational involvement at all.
• Comfort with pooled custody and the vault's withdrawal terms.

When a non-custodial basket makes more sense

The basket is the better structure when custody is a hard constraint, when you want more than one strategy running at once, or when you want to be able to stop instantly.

It is also better when you care about attribution. In a vault you see one number. In an isolated basket you can see which leader produced which part of the result, which is the only basis on which a replacement decision can be made rationally.

It requires enough capital to fund several sleeves, and it accepts tracking error — your fills will not match the leader's — in exchange for that structure. Those are real costs, not footnotes.

Risk that neither structure removes

Both structures are exposure to leveraged perpetual futures traded by someone else. Both can lose money quickly. Both are subject to liquidation, funding costs, exchange risk and the possibility that a leader's past record simply does not repeat.

Diversification across leaders reduces the impact of any one of them being wrong. It does not help when everything is wrong at the same time, which is exactly what happens in a correlated deleveraging event.

Past performance is not indicative of future results. Perpetual futures are leveraged instruments and carry a substantial risk of loss, including the loss of your entire position.

Conclusion

The question is not which structure is better in the abstract. It is which set of trade-offs you are willing to hold: pooled custody and simplicity, or self-custody and operational surface; one strategy cleanly executed, or several strategies approximately mirrored and separately measurable; a profit share, or a volume fee.

If you want to see how the basket structure is implemented in detail — the permission scope, the isolation model, the scoring and the replacement rules — the mechanics are written out on the how-it-works page, and the residual risks are stated plainly in the risk disclosure.