How to get Hyperliquid historical data — for free

What's in the archive

Reservoir covers the entirety of Hyperliquid activity, delivered as parquet:

• Fills — every trade, liquidation, ADL, builder fill, and TWAP fill, across Hyperliquid and every HIP-3 deployer.
• Candles — 1-second OHLCV. Backfilled before the August 2025 cutoff, so candle history reaches further back than the rest of the archive.
• Snapshots — daily positions, spot balances, and account values.
• Orderbook — 20-level L2 snapshots at 1-minute cadence, for every venue we cover.

If it happened on Hyperliquid, it's in the bucket.

The layout is plain S3 prefixes:

s3://hydromancer-reservoir/
├── fills/hyperliquid/      # weekly parquet, one prefix per venue
├── fills/ventuals/
├── fills/markets-xyz/
├── candles/1s/             # weekly parquet
├── snapshots/daily/        # daily parquet
└── orderbook/L2-1m/        # weekly parquet

Exact keys and column definitions are in the schema reference: fills, candles, snapshots, orderbook.

How to download Hyperliquid data from S3

You need an AWS account with configured credentials — not because we gate anything, but because requester-pays needs someone to bill the bandwidth to.

1. List the bucket.

aws s3 ls s3://hydromancer-reservoir/ --request-payer requester

2. Browse a dataset.

aws s3 ls s3://hydromancer-reservoir/fills/hyperliquid/ --request-payer requester

3. Copy what you need.

aws s3 cp \
  s3://hydromancer-reservoir/fills/hyperliquid/<key>.parquet . \
  --request-payer requester

4. Or skip the download and read straight into a dataframe:

import pandas as pd  # pip install pandas s3fs pyarrow

fills = pd.read_parquet(
    "s3://hydromancer-reservoir/fills/hyperliquid/<key>.parquet",
    storage_options={"requester_pays": True},
)

boto3 works too — pass RequestPayer="requester" on get_object. There's nothing proprietary in the path: any S3-compatible tooling that understands requester-pays will do.

What does requester-pays actually cost?

You pay AWS, not us. Two line items:

• Requests — about $0.0004 per 1,000 GETs at standard AWS rates. Rounding error.
• Transfer — roughly $0.09/GB out of AWS to the internet. Zero between S3 and compute in the same region.

Parquet compresses well, so a typical backtest pulls gigabytes, not terabytes — single-digit dollars from a laptop. Pulling the entire archive? Run the job in the bucket's AWS region and transfer drops to zero; you're left paying request pennies. Region and layout details are in the docs, or ask in Discord.

Do you even need Reservoir? The official API, honestly

For plenty of jobs — no. The official Hyperliquid API is free and good. Live positions, the current book, funding rates, the last few hundred fills for a dashboard: use it and skip the rest of this article.

Where it stops working:

That's the whole pitch, and it's checkable: pull the same wallet from both sources and diff.

Why we made Hyperliquid historical data free

Two reasons. One idealistic, one self-interested. Both true.

The idealistic one. Hyperliquid's premise is that a perps exchange can run in the open. HIP-3 open interest crossed $1.6 billion within its first year; weekend oil markets made Bloomberg and the WSJ. Claims that size shouldn't require trusting anyone's dashboard — not the exchange's, not ours. With Reservoir, any researcher can check how organic the volume is, what user retention looks like, what the median trader's PnL actually is — and publish, with no paywall between them and the receipts. Onchain perps data should be verifiable by default.

The self-interested one. Reservoir is proof of work. Our paying customers — market makers, HIP-3 deployers, the biggest apps on Hyperliquid — run on our real-time streams and low-latency infrastructure. The free archive lets anyone audit the quality of the underlying data before ever talking to us: diff it against the chain, hunt for gaps, stress the schemas. Teams that outgrow weekly batches know where to find us. That's the business model, stated plainly.

What Reservoir doesn't have

Worth being precise, so you don't architect around the wrong assumptions:

• No real-time. Fills, candles, and orderbook partitions land weekly; snapshots land daily. If your strategy needs the live book, that's our streaming stack, not the archive.
• L2 at 1-minute cadence. The free archive gives 20 levels every minute. Block-granular orderbook reconstruction exists, but on the enterprise side — talk to us.
• Hyperliquid crypto perps start August 2025. That's the archive cutoff. Candles are backfilled before it; fills, snapshots, and orderbook aren't. Every HIP-3 deployer is complete from its first trade.
• No SQL. It's S3 plus parquet, not a hosted query engine. Bring your own compute — DuckDB on a laptop goes surprisingly far.

Missing a dataset? We add to Reservoir based on demand. Ask in Discord.

What people build with it

• Backtests — signal research on 1-second candles, validated against actual fills, liquidations and ADLs included. Full history from launch, nothing held back.
• PnL and retention research — daily position and balance snapshots make cohort analysis straightforward: who traded first on HIP-3, who stayed, who's up.
• Builder analytics — builder fills and TWAP fills power attribution dashboards and revenue accounting for HIP-3 deployers.
• Microstructure studies — 20-level depth across every venue, for spread, depth, and impact work.

If you publish something built on Reservoir data, post it in Discord. We share the good ones.

When the archive isn't enough

Two layers sit on top of Reservoir, behind one API key:

• On-demand REST — per-wallet history in a single call: userFillsByTime, builderFillsByTime, userFunding, historicalOrders, and every other Hyperliquid historical endpoint — with time-range support and no fills limit.
• Real-time streams — websocket orderbook feeds, low-latency market data, Trade[XYZ] history, HIP-4 prediction markets, and the full HIP-3 stack. This is what 80+ apps, market makers, and trading teams run on.

Just starting out? You may be eligible for the early-stage teams program 🌱

FAQ