Hyperliquid data sources compared
Four ways to get Hyperliquid data. The info API is free but per wallet and capped. The S3 archive goes back further, but you pay the transfer and parse the files. Your own node records only from the day it starts. A hosted data API such as Ironflow has every fill since 25 May 2025 behind one REST call.
- API
- GET https://api.ironflow.sh/v1/trades?market=BTC-PERP no key
- SQL
- POST https://api.ironflow.sh/v1/query
- MCP
- https://mcp.ironflow.sh/mcp
Free key14 days of history at 60 requests a minute. Without a key, the API reaches 24 hours back.
By requesting a key you agree to the Terms and API terms; see the Privacy Policy. We may email you when your key keeps hitting its limits; you can opt out at any time.
The short version#
- Info API: free, live, per wallet. 5,000 candles and 10,000 fills at most.
- S3 archive: every fill by block and book snapshots. Requester pays, files arrive about monthly.
- Own node: everything, from the day it starts. A large server and daily upkeep.
- Hosted API: full history over REST, SQL and MCP. Free for 24 hours with no key.
Side by side#
| Source | Coverage | History | Cost | Effort |
|---|---|---|---|---|
| Info API | Live state, books, candles, funding. Fills and orders one wallet at a time. No liquidations request. | 5,000 candles per interval. A wallet's 10,000 most recent fills. | Free. 1,200 weight a minute per IP. | Low |
| S3 archive | Every fill by block, L2 book snapshots, asset contexts, transfers and funding events. No candles. | Fills in node_fills_by_block, older periods in node_fills and node_trades. Uploads about monthly, with no guarantee. | You pay AWS transfer. | High: AWS, LZ4 files, your own parser and storage |
| Your own node | Everything the chain produces, from the moment it runs. | None before it starts. | A 16 vCPU, 128 GB RAM, 500 GB SSD server. | Highest: about 100 GB of logs a day, peers, upgrades |
| Hosted data API (Ironflow) | Every fill, trade and liquidation, candles 1m to 1w, funding, open interest, prices, transfers. REST, SQL, streams, MCP. | Every fill since 25 May 2025, over 4 billion. | Free to start. $49 or $149 for 30 days, or from $0.001 a call. | One HTTP call |
Info API#
POST https://api.hyperliquid.xyz/info. Right for current state: mids, books, asset contexts with mark price, funding and open interest, and one account's positions, orders and fills.
candleSnapshot: only the most recent 5,000 candles. About 3.5 days at 1m, 208 days at 1h.userFillsByTime: 2,000 per response, and only a wallet's 10,000 most recent fills.- No request returns liquidations. They are a field on fills.
- 1,200 weight a minute per IP. Most info requests weigh 20 (rate limits).
S3 archive#
Two requester-pays buckets, described on Hyperliquid's historical data page:
hyperliquid-archive: L2 book snapshots and asset contexts, uploaded about once a month. Hyperliquid says there is no guarantee of timely updates and data may be missing.hl-mainnet-node-data:node_fills_by_block(every fill, one line per block), older periods innode_fillsandnode_trades,misc_events_by_block(transfers, staking, funding), explorer blocks and L1 transactions.- No candles. You need an AWS account, LZ4, a parser and somewhere to keep the rows.
Your own node#
A non-validator node streams what the chain produces while it runs. Hyperliquid's node README asks for 16 vCPUs, 128 GB RAM and a 500 GB SSD on Ubuntu 24.04, in Tokyo for the lowest latency.
--write-fills,--write-order-statuses,--write-raw-book-diffsand--write-misc-eventswrite hourly files.- No history: the local info server answers no historical time series.
- About 100 GB of logs a day to archive or delete. Our node guide covers setup and the errors we hit.
Hosted data API#
Ironflow runs its own nodes and backfilled from the archive: over 4 billion fills since 25 May 2025, both sides of each trade. In the last day alone, 14.5 million fills and 7.5 million trades.
curl "https://api.ironflow.sh/v1/trades?market=BTC-PERP&limit=3"- No key: 30 requests a minute, 300 a day, 24 hours of history.
- Free key: 60 a minute and 14 days. Get one by email.
- Paid: Explorer, $49, reaches 90 days back and Builder, $149, a year. Each payment covers 30 days. Enterprise has everything.
- No account: x402 pay-per-call in USDC on Base, $0.001 to $0.01 a call, with a year of history.
- Agents: 54 tools on
https://mcp.ironflow.sh/mcp, and read-only SQL over the same tables. - Free files: hourly candles, funding and liquidations as Parquet datasets.
Which one for which job#
- Live prices and one account's current state: the info API.
- Candles older than 5,000 bars: a hosted API, or free hourly files.
- A wallet's fills past the 10,000 cap: a hosted API, or the S3 archive filtered by address.
- Liquidations for every wallet: a hosted API, or the S3 archive filtered on the liquidation field.
- Raw book diffs and the lowest latency: your own node in Tokyo.
Questions#
What is the best source of Hyperliquid data?#
It depends on the job. The info API is free and right for live state and your own account. History past its caps comes from the S3 archive, your own node from the day it starts, or a hosted data API. Ironflow's API has every fill since 25 May 2025 behind one REST call, with no key for the last 24 hours.
Is Hyperliquid historical data free?#
The info API is free but capped at 5,000 candles per interval and 10,000 fills per wallet. The S3 archive is requester-pays, so you pay the AWS transfer. Ironflow publishes free hourly candles, funding and liquidations as Parquet files and serves 24 hours over its API with no key.
Does running a Hyperliquid node give me historical data?#
No. A node records from the moment it runs, with --write-fills and the other write flags, and its local info server answers no historical time series. Hyperliquid recommends 16 vCPUs, 128 GB RAM and a 500 GB SSD for a non-validator, and the node writes about 100 GB of logs a day.
What is in Hyperliquid's S3 archive?#
L2 book snapshots and asset contexts in hyperliquid-archive, uploaded about once a month with no guarantee. Fills by block, misc events, explorer blocks and L1 transactions in hl-mainnet-node-data. No candles. The requester pays the transfer.
Related guides#
- Hyperliquid historical data: API, S3 archive, SQL and free downloads
- Hyperliquid userFills limit: get more than 10,000 fills
- Hyperliquid historical liquidations API
- How to run a Hyperliquid non-validator node
References#
- Hyperliquid docs: Info endpoint. Request types and their limits.
- Hyperliquid docs: Rate limits. Weights per IP.
- Hyperliquid docs: Historical data. The S3 buckets and what each holds.
- hyperliquid-dex/node. Machine specs, write flags and the local info server.
Try the hosted route with no signup: curl https://api.ironflow.sh/v1/liquidations/summary.
Free key14 days of history at 60 requests a minute. Without a key, the API reaches 24 hours back.
By requesting a key you agree to the Terms and API terms; see the Privacy Policy. We may email you when your key keeps hitting its limits; you can opt out at any time.