hypercopy vs building your own hyperliquid copy trading bot

Github has a dozen hyperliquid copy-trading bots in python and node. They poll or subscribe to a wallet, compute position deltas and place orders. Some are decent. If you can run one, you pay no builder fee and trust nobody. This page is the honest account of what the rest of the job is, because that is where the cost moves.

what a basic copier does

• Watch a target address (websocket on the public trade feed, or polling its positions).
• When its position changes, compute the delta.
• Scale by a ratio and send a market-like order.
• Log.

A weekend of work. It will copy a swing trader on BTC reasonably. Then the details arrive.

the details

Detection. Hyperliquid's public trades feed carries both counterparties of every fill, so one subscription per coin covers every trader. But the first message on each subscription is a snapshot of old trades that must be ignored, subscriptions are capped per ip, and the feed disconnects. Polling positions instead is simpler and seconds slower.

Coalescing. A $200,000 order fills as forty pieces over two seconds. Naively, that is forty orders in your account, each paying fees and the minimum. You need to group fills within a window into one move.

Minimums and rounding. $10 minimum notional, per-coin size decimals, price tick sizes. Partial reductions below the minimum need a rule (skip, or round up to a full close).

Exits. Reduce-only on every reduction, or a rounding error flips you short. Full close when the trader is at 99.5%. Flip handling: close, then open the other side.

Idempotence. A retry after a timeout must not send the order twice. Client order ids solve it, once you have them everywhere.

Slippage. Hyperliquid has no market order. Immediate-or-cancel at mid ±x%, and a decision for when it does not fill. For exits, retry until flat.

Reconciliation. Missed a websocket message during a reconnect? Your position now differs from the trader's. A periodic pass comparing both and closing what the trader no longer holds.

Risk controls. A cap per trade, a stop-loss on margin checked on live prices, a circuit breaker after repeated rejections, a limit on insufficient-margin retries.

Keys. An agent key, valid at most 180 days, never reused (nonce replay), stored encrypted, rotated before expiry with the positions handled.

Hosting. A server close to hyperliquid's api for consistent latency, a static ip, process supervision, alerting when it stops.

Discovery. The leaderboard, then filters for account size, market makers, vaults, taker ratio and hold time, or you will spend your first month copying a market maker.

Hypercopy is that list, running, with a builder fee of 0.05% per order as the price. See copy latency and slippage for the parts that decide fill quality.

when building your own is right

• You are copying one trader you know, on liquid pairs, and can live with a few seconds of delay from polling.
• You need something no product does: multi-venue hedging, custom sizing logic, copying into a sub-account structure.
• You are a developer and you want to understand the mechanics. It is a good project.
• You trade enough size that 0.05% per order pays for a server and your time.

when it is not

• You want it to run while you sleep, and you are not the person who wants to be paged at 4 am.
• You want to copy several traders with different sizes and stop-losses.
• You have less than the time and size to make the fee saving matter.

faq

can i use hypercopy's approach with my own bot?

The architecture is described in the pages linked here. The code is not open source.

will an open-source bot see my private key?

It runs on your machine with an agent key you generate. That is fine as long as the machine is yours and the key is an agent, not your wallet key. See agent keys explained.

is polling positions good enough?

For swing traders on liquid pairs, often yes. For anything faster, the fill feed is the difference between copying and arriving late.