Non-custodial copy trading on Hyperliquid, explained

Custody versus permission

Custody is the ability to move funds. Permission is the ability to act within an account without moving funds out of it. Conflating the two is why the word 'non-custodial' has become unreliable: a platform can honestly say it never touches your withdrawal rights while still holding a permission broad enough to cause real damage, and another can call itself non-custodial while requiring a deposit into a pooled address.

The test is mechanical, not rhetorical. Ask which key signs a withdrawal. If the answer is any key other than yours, the arrangement is custodial regardless of the language used to describe it.

What a Hyperliquid agent approval actually grants

Hyperliquid supports agent wallets: you sign an approval that names a specific agent address and authorizes it to submit trading actions for your account. That agent key can open, size, modify and close positions. It is the same primitive whether it is used by your own script or by a copy system.

The approval is scoped to trading. It does not confer ownership, and it does not extend to the account's fund-movement actions. It is also revocable: you can withdraw the permission with your own wallet, and doing so does not require the operator's cooperation or any transfer of assets.

• Grants: placing, modifying and closing perpetual-futures orders on your account.
• Does not grant: withdrawals, transfers to other addresses, or account ownership changes.
• Revocable on-chain by the account owner at any time.

The builder-fee approval, and why it is separate

Hyperliquid has a native builder-fee mechanism: the party that builds an order can attach a fee, and the account owner approves a maximum rate in advance. This is a second, distinct signature from the agent approval, and it exists so the fee is bounded by something you signed rather than by a promise in a pricing page.

HyperMirror charges 0.1% of mirrored notional volume through this mechanism. Because the charge rides on the order itself, it is visible on-chain per fill, and because the maximum rate is approved by you, it cannot be raised unilaterally beyond what you authorized. There is no subscription and no share of your equity.

Sub-accounts: isolation without giving up custody

Hyperliquid sub-accounts belong to your main account. They let capital be partitioned into separate margin and position contexts while remaining under the same ownership, which is what makes multi-leader mirroring possible without custody and without netting.

HyperMirror uses one sub-account per mirrored leader. Two leaders holding opposite sides of the same perp keep both positions instead of cancelling into a single net exposure; a margin problem in one sleeve does not consume the margin backing another; and per-leader results stay attributable, which is what scoring and replacement depend on.

None of that changes who owns the funds. The sub-accounts are yours, and their balances are visible to you on-chain at all times.

Contrast: custodial CEX copy trading

On a centralised exchange, copy trading generally requires your funds to sit in an account the exchange controls, sometimes routed through an internal copy-trading wallet. Your position is a database entry that the exchange can freeze, and your ability to exit depends on the exchange processing a withdrawal.

That structure adds a risk that has nothing to do with trading: the solvency and operational integrity of the custodian. It has been the dominant source of total loss in this industry, and it is not diversifiable by picking better traders.

Contrast: API-key bots

API-key bots look non-custodial and often are not, because API permissions are coarse. Keys are frequently issued with withdrawal rights available, and the security of the arrangement then depends on the user setting the right toggles and the operator storing the key safely. A leaked key with withdrawal enabled is equivalent to a stolen account.

An agent approval differs in that the narrowness is enforced by the exchange's action model rather than by configuration hygiene. The agent key simply cannot sign a withdrawal, so a compromise of that key is bounded to trading damage — serious, but not total loss of the balance.

Contrast: vaults and pooled structures

A vault pools depositors' capital into a single managed account. Your exposure is a share of a pool rather than positions you own, exit is a redemption subject to the vault's terms, and every depositor's positions net together inside one book.

Vaults are a legitimate structure with real advantages — simplicity, and a manager who can act on the whole book at once. But they are not non-custodial in the sense used here, and the difference shows up precisely when you most want to leave.

What non-custodial does not protect you from

Being non-custodial removes counterparty risk on your balance. It does not remove market risk, and it is important not to let the custody story do work it cannot do. A trade-only agent can still open leveraged positions that lose money, and a mirrored leader can still be liquidated.

It also does not eliminate execution divergence: your fills, timing and account size differ from the leader's, so your results will differ from theirs. And it does not remove operational risk in the software itself, which is why per-leader notional ceilings, independent leverage caps and sub-account isolation matter as bounds on what a malfunction can cost.

Nothing here is financial advice. Perpetual futures are leveraged instruments: a position can be liquidated in full, and past performance of any trader is not indicative of future results. Copy trading does not remove that risk — it changes who makes the decision, not what the market can do to it.