Skip to main content

Where to get Hyperliquid historical data: S3, the API, a node, or a provider

Hyperliquid historical data comes from four places: Hyperliquid's S3 buckets, the info API, a node you run, or a data provider. How far back each one goes and what each one misses.

14 min read
Explore with your agent
You needFree sourceWhat it misses
A chart of recent pricesInfo API candlesAnything older than 5,000 bars
Funding historyInfo API fundingSpeed, 500 rows a call
Order book historyS3 L2 snapshotsMonthly uploads and gaps
Open interest historyS3 asset contextsMonthly uploads
Every trade, market-wideS3 node fillsAnything before March 2025 in the API format
Your own tradesInfo API fillsFills past the latest 10,000
Every order, from today onYour own nodeEverything before the day you start it

Key findings

  1. Hyperliquid's S3 archive has L2 snapshots and fills, uploaded about once a month
  2. The info API gives candles and funding, but only recent history
  3. A node records everything from the day you start it, and nothing before
  4. GitHub download tools pull from the same S3 buckets
  5. Which Hyperliquid data source fits which job
  6. Other sources and where they stop

Short answer: to download Hyperliquid historical data without an AWS account or your own node, use 0xArchive. For BTC it serves order books from 15 April 2023, trades from 19 April 2023, funding and open interest from 20 May 2023 and liquidations from 22 December 2025, through the REST API, WebSocket replay and Parquet exports. The last 30 days are free, and Build opens the full archive. Hyperliquid's S3 archive suits raw L2 snapshots if you can pay AWS transfer costs and wait for monthly uploads.

0xArchive market data guide · Hyperliquid

Hyperliquid historical data comes from four places. Hyperliquid's own S3 buckets, the public info API, a node you run yourself, and data providers that were recording before you started.

Each one covers a different slice of history. Pick the wrong one and you find out halfway through a backtest.

Bar chart of how far back each Hyperliquid data source reaches. S3 L2 snapshots from April 2023, S3 fill files from March 2025, info API candles 208 days at 1 hour and 3.5 days at 1 minute, a new node from today, 0xArchive from April 2023.
The span of history each source serves, on 7 October 2026 UTC. Info API candle spans are the most recent 5,000 bars at each interval. The S3 archive is uploaded about once a month, so its newest weeks arrive late. 0xArchive start dates are the first day each data type is served for Hyperliquid perps. Get this data

Hyperliquid's S3 archive has L2 snapshots and fills, uploaded about once a month

Hyperliquid publishes raw history to two Amazon S3 buckets, documented on its historical data page. You need an AWS account to read them, because the requester pays the transfer cost.

BucketWhat is in itFormat
hyperliquid-archiveL2 book snapshots (market_data) and asset contexts (asset_ctxs)LZ4 files, hourly per coin
hl-mainnet-node-dataFills by block, explorer blocks, L1 transactions, transfers, staking and funding eventsRaw node output

The asset contexts file is where historical open interest, funding and mark prices live, one CSV per day.

The L2 snapshots go back to April 2023. Fills in the same format as the API start in March 2025, and older trade files use a different format.

aws s3 cp s3://hyperliquid-archive/market_data/20230916/9/l2Book/SOL.lz4 /tmp/SOL.lz4 --request-payer requester
unlz4 --rm /tmp/SOL.lz4

Hyperliquid's docs say uploads happen "approximately once a month", and "there is no guarantee of timely updates and data may be missing". There are no candles and no spot data in the archive.

So S3 suits a research project that can wait for last month's files and is fine parsing raw node output. It does not suit anything that needs yesterday.

The info API gives candles and funding, but only recent history

The public info API at api.hyperliquid.xyz is free and needs no account. It answers POST requests for candles, funding, the current book and your own account.

Candles are capped. Hyperliquid's docs put the limit at the most recent 5,000 candles for any coin and interval.

IntervalHow far back 5,000 candles reach
1 minute3.5 days
15 minutes52 days
1 hour208 days
1 dayHyperliquid's whole history
curl -s https://api.hyperliquid.xyz/info \
  -H 'Content-Type: application/json' \
  -d '{"type":"candleSnapshot","req":{"coin":"BTC","interval":"1m","startTime":1681992000000}}'

Ask for 1-minute bars from April 2023 like this and the response still starts 3.5 days ago. For the full one-minute history, see how to download Hyperliquid OHLCV candles.

Funding history goes back further. Responses return 500 rows at a time, and you page forward by timestamp.

The rest is current state only. The L2 book endpoint returns the book right now, there is no endpoint for past open interest, and there is no market-wide trade history at all. Your own fills stop at the 10,000 most recent.

A node records everything from the day you start it, and nothing before

A non-validating node sees everything the chain does. Run with the right flags, it writes every fill, every order status and every raw book change to disk as blocks arrive.

The node writes from the moment it is running. Hyperliquid's node docs say its local server does not answer historical time series queries, so a node started today has no yesterday.

Hyperliquid's node docs list the non-validator minimum:

ResourceMinimum
CPU16 vCPUs
Memory128 GB RAM
Storage500 GB SSD, Ubuntu 24.04

Storage fills fast. With default settings the node writes about 100 GB of logs a day, and order statuses alone run several hundred GB a day.

A node also needs a steady connection to its peers. The Hyperliquid Foundation lists infra providers for non-validating nodes on its docs, and 0xArchive is one of them, with dedicated Tokyo peering at $200 per node per month. See the setup instructions, or use the peering migration guide if your node already peers elsewhere.

GitHub download tools pull from the same S3 buckets

Searching GitHub for Hyperliquid historical data turns up download scripts, not new data. The most used one, hyperliquid-historical, downloads the hourly L2 files from the hyperliquid-archive bucket, decompresses them and converts them to CSV.

These tools save you the AWS command line. They still need an AWS account, they still pay the requester's transfer cost, and they carry every gap the S3 archive has.

Which Hyperliquid data source fits which job

JobFree sourceWhat it misses
A chart of recent pricesInfo API candlesAnything older than 5,000 bars
Funding historyInfo API fundingSpeed, 500 rows a call
Order book historyS3 L2 snapshotsMonthly uploads and gaps
Open interest historyS3 asset contextsMonthly uploads
Every trade, market-wideS3 node fillsAnything before March 2025 in the API format
Your own tradesInfo API fillsFills past the latest 10,000
Every order, from today onYour own nodeEverything before the day you start it

Other sources and where they stop

The S3 archive, the info API and a node are the three places Hyperliquid itself provides. Several outside providers also publish Hyperliquid history, each with a different start date and a different shape. This is where each one begins and where it ends.

SourceHyperliquid data it holdsHistory fromWhere it stops
Hyperliquid S3L2 book snapshots, asset contexts, fills, node data; requester pays for transferBook snapshots April 2023, fills in the API format March 2025Uploaded about once a month with no guarantee of timely updates; no candles or spot asset data
Hyperliquid info APICandles, funding, the current book, your own fillsLast 5,000 candles per intervalRate limit of 1,200 weight a minute per IP; no market-wide trade history
Your own nodeEvery fill, order status and raw book changeThe day you start it16 vCPUs, 128 GB RAM and a 500 GB SSD at the minimum; about 100 GB of logs a day by default
Hydromancer ReservoirFills with liquidations, ADLs, builder and TWAP fills; 1-second candles; daily positions; 20-level L2 snapshots once a minute; requester-pays S3 Parquet, no API keyFills 28 Jul 2025; candles 31 Jul 2025; books 15 Dec 2025Nothing before 28 Jul 2025; book snapshots once a minute at 20 levels; no resting-order book listed
IronflowTrades, fills, liquidations, funding, candles, open interest; REST, SQL and MCP with a keyless tierFills since May 2025History follows the plan: 24 hours with no key, 30 days Free, 90 days at $49 per 30 days, 1 year at $149 per 30 days
VikeFills and open-position snapshots as daily Parquet files, plus REST and MCPFills 1 May 2026; positions 8 May 2026Nothing before May 2026; Free is 15,000 requests a month, the Parquet files are for non-commercial use, and the $49 plan is listed as opening soon
TardisRaw Hyperliquid WebSocket messages: 20-level book snapshots about every 5.4 seconds, trades with buyer and seller addresses, asset contexts29 Oct 2024; trade addresses from 1 Dec 2024; HIP-3 from 2 Dec 2025Free CSV covers the first day of each month; the raw replay API needs Professional or Business; yearly billing reaches back 4 years, monthly 4 months
AlliumFills, orders, transfers and metrics for Hyperliquid, delivered through Snowflake, Databricks and Google Cloud, with Kafka and Pub/Sub streamsTrades backfilled; orders from when Allium began running its own nodesAbout 4,000 traders with more than 10,000 trades at the March 2025 backfill have only their last 10,000 trades before March 2025; orders cannot be backfilled
CryptoHFTDataL2 order book images, tick trades, open interest and funding for spot and perpetual markets, as filesNot listed on its Hyperliquid pagePrice-level books only, no individual orders or wallets
BitqueryParquet exports of fills, order updates, book updates, funding and oracle prices; free samples of a few hundred rowsLast 30 days on its Data StoreOlder ranges only through its sales team; its own docs list hour-long gaps in September and October 2026
GoldRushHyperCore fills, orders, trades and funding through an Info API replacement, plus a live L4 book stream25 May 2025No Lighter; free use is a 14-day trial with 25,000 credits

Provider facts from each provider's own documentation and pricing, 6 November 2026.

Hyperliquid's info API. The rate limit is a weight budget shared by all REST requests from one IP address: 1,200 a minute. An l2Book request weighs 2, most other info requests weigh 20, and funding history and recent trades add weight for every 20 rows returned. Hyperliquid's own docs send large batch requests to the S3 bucket instead.

Running a node. Hyperliquid recommends Tokyo for the lowest latency. The node writes from the first block it applies, which is why the table says the day you start it.

Hydromancer Reservoir. The bucket is s3://hydromancer-reservoir, with Parquet in date=YYYY-MM-DD partitions, read directly by DuckDB with the region set to ap-northeast-1 (Tokyo). Hydromancer's page gives sizes measured on 18 October 2026: perp fills about 360 MB a day, 1-second candles about 44 MB a day and liquidation fills about 1 MB a day. Those sizes set the transfer bill, and AWS sets the price per GB. Like Hyperliquid's bucket, Reservoir is a set of files with no query API, so a missing day is a missing file and you reconcile formats yourself.

Ironflow, Vike, Tardis and Allium. Ironflow runs its own nodes and sells dedicated nodes from $1,250 a month. Vike's Hyperliquid files are fills and position snapshots. Tardis records the exchange's public WebSocket feed, so its books are the snapshots Hyperliquid publishes, and it also covers Binance, OKX, Coinbase, Bybit and other exchanges. Allium is an onchain data platform for institutions across 20+ chains, built for stablecoin, payments and cross-chain analytics, with Hyperliquid as one of its chains.

What 0xArchive covers that those sources do not

Everything below is served by symbol and UTC window from the same REST API, WebSocket and Parquet exports.

  • Order books from 15 April 2023. Native L2 with up to 20 levels per side, read at any timestamp. Early books carry about 10 levels.
  • L4 from 11 March 2026. Every resting order with its order ID and wallet, plus L4 diffs, order events and full-depth L2, on core Hyperliquid and HIP-3. Spot L4 starts 5 May 2026.
  • Trades, funding and liquidations. Trades from April 2023 and every fill from 22 March 2025. Funding and open interest from 20 May 2023. Liquidation events with the liquidated and liquidator wallets from 22 December 2025.
  • HIP-3 and HIP-4. HIP-3 liquidations and candles from 22 December 2025. HIP-4 from 2 May 2026, without funding or liquidations.
  • Lighter. Trades from 17 January 2025, order books from 29 January 2026, and L3 individual resting orders from 5 March 2026, attributed by Lighter account index rather than wallet. Lighter L3 is history and replay, not live.
  • Live WebSocket, replay and Parquet. One schema for live feeds, historical replay and file exports, plus an MCP server and SDKs.
  • Plans. Free is $0 with every route and WebSocket channel except the live mempool stream, the latest rolling 30 days, live WebSocket and replay, 50,000 credits a month and 15 requests a second. Build is $49 a month ($39 billed annually) with the full retained archive, 80 million credits and $50 in export credits each month. Trades cost 1 credit per 1,000 rows and L4 costs 1 credit per 5,000 rows, rounded up per request. Data is licensed for your own use, with no redistribution.

To move a file workflow onto the API, replace each file read with one request:

You read todayRequest instead
market_data/[date]/[hour]/l2Book/[coin].lz4GET /v1/hyperliquid/orderbook/{coin}?timestamp=... for one moment, or /orderbook/{coin}/history?start=...&end=... for a window
asset_ctxs/[date].csv.lz4GET /v1/hyperliquid/openinterest/{coin} and /funding/{coin}
node_fills_by_block, or Reservoir fills/perp/allGET /v1/hyperliquid/trades/{coin}?start=...&end=..., then follow next_cursor
Reservoir fills/perp/liquidationsGET /v1/hyperliquid/liquidations/{coin}, events from 22 Dec 2025 with both wallets
Reservoir candles/1sGET /v1/hyperliquid/candles/{coin}?interval=1m, 1 minute to 1 week from 22 Mar 2025, or build any bar size from trades
Reservoir orderbook/1mThe book at any timestamp instead of one snapshot a minute: native L2 from 15 Apr 2023, full-depth L2 and L4 from 11 Mar 2026

Whole months as files come from the Data Catalog as Parquet. For book grain, see Hyperliquid order book historical data and the full provider comparison. Start free with the latest 30 days, or see plans and export pricing.

Full Hyperliquid history from one API with 0xArchive

0xArchive records Hyperliquid and serves it from one REST API, WebSocket replay and Parquet file exports. The history you would otherwise stitch together from S3, the API and a node comes back from the same routes, in UTC, with cursors.

Hyperliquid dataServed from (UTC)
L2 order bookApril 2023
Funding and open interestMay 2023
CandlesMarch 2025
LiquidationsDecember 22, 2025
L4 orders, every resting order with its walletMarch 2026
  1. April 2023L2 order book served from this month.
  2. May 2023Funding and open interest.
  3. March 2025Candles.
  4. March 2026L4 orders, every resting order with its wallet.
curl "https://api.0xarchive.io/v1/hyperliquid/orderbook/BTC/history?start=1681992000000&end=1681992060000&limit=5" \
  -H "X-API-Key: $OXARCHIVE_API_KEY"

That returns the BTC book from 20 April 2023, the same window the info API cannot reach. The venue coverage page lists the first served instant for every data type.

Reproduce the archive example

The April 20, 2023 request uses full-archive access, available on Build and above. To try the same BTC order-book route with a Free key, use the recent example below. To inspect the data without a key, download the fixed BTC Parquet sample.

Try a recent BTC order book with a Free key

Use the same native L2 history route with a recent, bounded UTC window. Free covers the most recent rolling 30 days. Create a free key, then save the resolved start and end with the response so another researcher can repeat the request while that window remains in their plan. This example resolves a one-minute window ending one minute ago and returns the first page only; follow meta.next_cursor for the rest of the window.

import json
import os
import time
import urllib.error
import urllib.parse
import urllib.request
end_ms = (int(time.time() * 1000) // 60000 - 1) * 60000
start_ms = end_ms - 60000
query = urllib.parse.urlencode({"start": start_ms, "end": end_ms,
                              "limit": 100, "depth": 10})
request = urllib.request.Request(
    "https://api.0xarchive.io/v1/hyperliquid/orderbook/BTC/history?" + query,
    headers={"X-API-Key": os.environ["OXARCHIVE_API_KEY"]})
receipt = {"requested_start_ms": start_ms, "requested_end_ms": end_ms}
try:
    with urllib.request.urlopen(request, timeout=30) as response:
        receipt["status"] = response.status
        receipt["response"] = json.load(response)
except urllib.error.HTTPError as error:
    receipt["status"] = error.code
    receipt["error_body"] = error.read().decode("utf-8", "replace")
print(json.dumps(receipt))

Download a BTC order-book sample (.parquet)

Download the fixed BTC sample. It contains 20 recorded snapshots from August 10, 2026, not a full day of continuous book updates. Inspect its JSON preview and published manifest.

Read the downloaded sample with DuckDB:

DESCRIBE SELECT * FROM read_parquet('0xarchive-hyperliquid-btc-orderbook-v1.parquet');
SELECT timestamp, bids, asks
FROM read_parquet('0xarchive-hyperliquid-btc-orderbook-v1.parquet')
ORDER BY timestamp;

For a flat CSV, download the BTC trades Parquet and run the DuckDB command below. The order-book file preserves bids and asks as nested levels.

COPY (SELECT * FROM read_parquet('0xarchive-hyperliquid-btc-trades-v1.parquet'))
TO 'btc-trades.csv' (HEADER, DELIMITER ',');

Both samples are 0xArchive Data under the Data License.

Need a different source? Compare data providers.

Common questions about Hyperliquid historical data

Is Hyperliquid historical data free?

Yes, in pieces. The info API is free, and the S3 archive is free to read apart from the AWS transfer cost the requester pays.

How far back do Hyperliquid API candles go?

The info API serves the most recent 5,000 candles per coin and interval. That is 3.5 days of 1-minute candles or 208 days of hourly ones.

Where does Hyperliquid store historical data?

In two S3 buckets, hyperliquid-archive for L2 book snapshots and asset contexts, and hl-mainnet-node-data for fills and node data. Hyperliquid uploads them about once a month.

Can a Hyperliquid node give me past data?

No. A node writes data from the moment it starts running, so history before that date has to come from S3 or a data provider.

Is there free Hyperliquid historical data with order books and wallets?

Yes. 0xArchive Free is $0 and covers every route and WebSocket channel except the live mempool stream for the latest rolling 30 days, including L4 with order ID and wallet, with 50,000 credits a month and 15 requests per second. Live WebSocket and replay are included.

How do I get Hyperliquid data older than 30 days?

Build is $49 a month ($39 billed annually) and opens the full retained archive: order books from 15 April 2023, trades from April 2023, every fill from 22 March 2025, funding and open interest from 20 May 2023, liquidation events from 22 December 2025 and L4 from 11 March 2026.

Run this analysis from an agent

Hyperliquid books, trades, funding and open interest all come from routes an agent can call. 0xArchive ships an MCP server, so Claude and other AI agents can pull Hyperliquid history in plain language. There is a Claude Code skill for terminal workflows and an agentic gateway for agents that pay per call. Start with the agent surfaces overview or the guide to replayable market data for MCP, CLI and agent workflows.

Route references

Use the order book routes reference for L2 and L4 history, the trades reference for fills, the funding reference and open interest reference for positioning, and the candles reference for bars. The guide to Hyperliquid historical data for backtesting covers pulling a window cleanly, and the one-symbol parity test shows how to check Hyperliquid's own data against managed history before you switch. For paid providers side by side, see the data provider comparisons.

For current API availability and plan details, use the Hyperliquid data API, data catalog, pricing, and status.