Hyperliquid's API is public and needs no key, which is why most projects start there. It also comes with limits, and the day a bot, a dashboard or a backfill hits one, the error rarely says which. This guide puts them in one place, quoted from Hyperliquid's documentation, then shows what a Hypedexer key changes.
REST: a weight budget per IP
Every IP gets one budget: "REST requests share an aggregated weight limit of 1200 per minute." Each request spends part of it:
| Request | Weight |
|---|---|
l2Book, allMids, clearinghouseState, orderStatus, spotClearinghouseState, exchangeStatus | 2 |
Every other documented info request | 20 |
| Explorer requests | 40 |
userRole | 60 |
| Exchange actions (orders, cancels) | 1 + floor(batch length / 40) |
Requests that return lists, such as userFills, userFillsByTime, historicalOrders, recentTrades and fundingHistory, also have "an additional rate limit weight per 20 items returned in the response". At a weight of 20, that is 60 info requests a minute, or one a second.
History: the most recent items only
The info endpoint is built for recent activity:
userFills"returns at most 2000 most recent fills".userFillsByTime"returns at most 2000 fills per response and only the 10000 most recent fills are available".historicalOrders"returns at most 2000 most recent historical orders".- "Responses that take a time range will only return 500 elements or distinct blocks of data", so longer ranges are paged by moving
startTimeforward.
For an active account, the 10,000 most recent fills can be a few weeks. Everything before that is still on-chain, but this API will not return it.
WebSocket limits
From the same page, per IP:
- 10 WebSocket connections, and 30 new connections a minute.
- 1,000 subscriptions.
- 10 unique users across user-specific subscriptions.
- 2,000 messages sent to Hyperliquid a minute, and 100 in-flight
postmessages.
The one people meet first is the third: a tracker that follows more than 10 wallets on user-specific subscriptions needs another way in.
When the limits bite
- Backfills. Rebuilding an account's year of trades is impossible past the 10,000 most recent fills, and slow before that at 2,000 per response.
- Many wallets. Copy-trading screens, risk desks and leaderboards follow hundreds of addresses, not ten.
- Shared IPs. Every process behind one address, every user of a hosted notebook, spends the same 1,200 a minute.
- The whole market. Liquidations, funding or fills across every account are not one call: the documented endpoints are per user.
The same data on a Hypedexer key
Hypedexer indexes Hyperliquid from its own nodes and serves it by date range, so history is a query, not a window. Limits come with the plan, not with the IP:
| Plan | Requests | WebSocket | Credits a month |
|---|---|---|---|
| Free | 1 a second | 1 connection, 2 subscriptions | 5,000 |
| Starter, $249 | 20 a second, bursts to 60 | 5 connections, 200 subscriptions | 500,000 |
| Growth, $899 | 50 a second, bursts to 150 | 20 connections, 600 subscriptions | 3,000,000 |
| Scale, $1,999 | 100 a second, bursts to 300 | 50 connections, 1,500 subscriptions | 12,500,000 |
What a call costs, from the rate limits guide:
- Every REST call costs at least 1 credit; paginated responses cost 1 credit per 25 rows.
- Errors are free: no credits are deducted for a status of 400 or more.
- On WebSocket, 10 server-to-client events cost 1 credit.
- Each response carries
X-Credit-CostandX-Credit-Balance, so a pipeline always knows where it stands.
A wallet's trades, back to January 2024, page by page:
curl -H "X-API-Key: $HYPEDEXER_KEY" \
"https://api.hypedexer.com/fills/user/0xYourWallet?start_time=2024-01-01T00:00:00Z&end_time=2024-02-01T00:00:00Z&limit=1000&order=ASC"Pass next_cursor back as cursor until has_more is false. Fills before March 2025 are served over the API on Scale and Enterprise; recent ones on every plan. Plans and the stake discount are on pricing.
FAQ
What is the Hyperliquid API rate limit? 1,200 weight a minute per IP for REST. Most info requests weigh 20; l2Book, allMids and a few others weigh 2.
How many WebSocket connections does Hyperliquid allow? 10 per IP, with 1,000 subscriptions and 10 unique users across user-specific subscriptions.
Why does userFillsByTime stop returning older fills? Only the 10,000 most recent fills of an account are available through it. Older ones need an indexer, such as the Data API or the free Trade Export.
Do failed requests cost credits on Hypedexer? No. A response with a status of 400 or more deducts nothing.