⚠︎ Risk warning: leveraged derivatives on unregulated platforms — you can lose everything you deposit. Not investment advice.
Hyperliquid API and Bot Trading: Limits, Wallets, Fees
Rules and numbers verified against Hyperliquid's official API docs on 2026-09-15. This page is for traders deciding whether and how to automate; it is not an integration tutorial. AI-client access (MCP) is covered in the AI-agent guide.
Hyperliquid's API is the same interface the web app uses, documented publicly with an official Python SDK and community Rust and TypeScript clients, plus CCXT integrations "in multiple languages". Because the order book is on-chain, the API's limits and latency behave differently from a centralised exchange's. The docs are unusually explicit about those differences.
How requests reach the chain
"API servers listen to updates from a node and maintain the blockchain state locally", serving REST and WebSocket and forwarding your transactions to a node, which gossips them into HyperBFT consensus; the server "responds to the original request with the execution response from the L1" once the transaction is in a committed block. End-to-end latency therefore "includes time to API server, mempool inclusion time, and time to commit (usually 2 blocks)"; the docs quote "~380ms end-to-end latency" for a cancel or post-only order while noting that sequencing has "far less noise" than that figure suggests. Two ordering rules matter for any bot: "cancels and ALO orders sent at time t will almost always execute before IOC and GTC orders sent at time t", and cancels should carry the "fast" flag (not usable for trigger orders).
API wallets
Bots sign with API wallets, "also known as agent wallets in the docs". "A master account can approve API wallets to sign on behalf of the master account or any of the sub-accounts", and "API wallets are only used to sign" — queries must use the real account address, or you get "an empty result". Each master account starts with 3 API wallets, plus 2 per sub-account. Nonces are tracked per signer, the 100 highest per address, and must fall within "(T - 2 days, T + 1 day)" of block time; the docs recommend one API wallet per trading process and per sub-account, batching orders every 0.1 seconds, and — "strongly" — never reusing a deregistered agent's address, because its nonce state "may be pruned" and old signed actions "can be replayed". A new account pays a one-time activation fee of "1 quote token (e.g., 1 USDC, 1 USDT)" on its first inbound transaction.
Rate limits
| Limit | Per the docs |
|---|---|
| REST weight per IP | 1,200 per minute; exchange actions weigh 1 + floor(batch_length / 40); most info requests weigh 20, l2Book / allMids / clearinghouseState 2, userRole 60 |
| Websockets per IP | 10 connections, 30 new per minute, 1,000 subscriptions, 10 unique users across user-specific subscriptions, 2,000 messages per minute, 100 inflight posts |
| EVM JSON-RPC | 100 requests per minute at rpc.hyperliquid.xyz/evm |
| Address-based | 1 request per 1 USDC traded cumulatively, with a 10,000-request starting buffer; once limited, one request every 10 seconds; cancels get extra headroom |
| Open orders | 1,000 by default plus one per 5M USDC of volume, capped at 5,000; beyond 1,000 open orders, reduce-only and trigger orders are rejected |
| Congestion | addresses limited to 2x their previous-day maker share of block space |
The address-based rule is the one that shapes strategy design: "1 request per 1 USDC traded
cumulatively since address inception" means, in the docs' words, that "with an order value of
100 USDC, this requires a fill rate of 1%". Batched requests count as one for the IP limit but
as n for the address limit. Websocket connections are closed after 60 seconds without a
server message; a ping keeps them open.
Priority fees (advanced)
Latency-sensitive users can pay to move up the queue in two separate ways, both burned in HYPE. Gossip (read) priority is a Dutch auction on a 3-minute schedule with "approximately 25 ms reduction in latency per auction slot" and a 0.1 HYPE minimum bid. Order (write) priority is a per-order rate on IOC or post-only batches, "approximately 45 ms reduction in end-to-end latency per 1 bp of priority fee paid" up to 8 bps, charged as a fraction of filled (IOC) or resting (ALO) notional from the undelegated staking balance. Cancels still go ahead of every immediately executable order regardless of fee. The docs frame both as reducing "the need for market makers to hyper-optimize their infrastructure".
Market making and builder codes
The market-making page is one paragraph: "There is no DMM program, special rebates / fees, or latency advantages." The published maker-rebate tiers on the fees page are the whole incentive. Third-party front-ends and bots that route orders for you can charge a builder fee only with your signed approval; see the builder-codes page. Native multi-sig accounts exist as "a built-in primitive on HyperCore" for teams that want several keys over one account, with up to 10 authorised users and a threshold; the docs warn multi-sig users "should not interact with the HyperEVM".
FAQ
Can an API wallet withdraw my funds?
The nonces page describes API wallets as signing "on behalf of" the account; which actions a given key may sign is defined in the exchange-endpoint docs, which this page does not reproduce. Treat any key with withdrawal authority as equivalent to the account itself.
Is there a testnet?
Yes: "you can make the same requests against testnet using the corresponding url", and a testnet faucet is documented under onboarding.
Does the API cost anything?
No API fee. Trading fees apply as usual, the 1 USDC activation applies to new accounts, and priority fees are optional.
Open Hyperliquid (opens in a new tab)
Referral link — 4% fee discount, disclosed on the referral page. See our methodology.