The archive that matches the feed.
Every trade, order, book delta, TWAP, and ledger event on Hyperliquid, stored with byte-exact decimals, full wallet attribution, and the same schema as the live streams. Built for backtesting, research, and ML.
1,652,195,587
order events archived · +725M+ / day
- Trades archived
- 14,516,809
- Book updates archived
- 1,000,273,954
- Markets covered
- 462+
Continuous since launch, no TTL · as of 2026-06-09
Six datasets. One row per event.
Not aggregates of the tape but the tape itself, down to individual order-status events and book deltas. Expand a row for the field-level detail.
Every dataset: exact decimals, full wallet attribution, no TTL. Field-level schemas →
Backtest on history, deploy on live, zero remapping.
The archive and the live gRPC streams are two views of one pipeline. Same fields, same names, same exact-decimal encoding. The loader you write for the backtest is the handler you ship to production. No translation layer, no “mostly compatible” surprises on deploy day.
Archive row · trades dataset
query API · Parquet / CSV / JSONL exports
Live message · StreamSwaps
stream.hyperliquidrpc.com:443 · gRPC
Hover a field. It is the same field, byte for byte, on both sides. Backtest on the archive, deploy on the stream, change nothing.
Byte-exact decimals. Never floats.
Most data vendors hand you float64 and call it precision. Floats cannot represent most decimal prices. The error is tiny per row and systematic across a backtest, which is the worst possible combination for anyone measuring edge in basis points.
We store every price, size, fee, and PnL figure as an exact decimal, end to end. The archive reconciles to the chain and to Hyperliquid’s public API. At 725M+ order events a day, “close enough” is not a storage format.
The same arithmetic, two storage models
float64 results reproducible in any IEEE-754 runtime
One fill: notional = px × sz
67012.0 × 0.14523
- float64drift
- 9732.152759999999
- archiveexact
- 9732.15276
One million 0.001 fees, summed
Σ 0.001 × 1,000,000
- float64drift
- 999.9999999832651
- archiveexact
- 1000.000
This is the arithmetic a backtest runs millions of times. The archive stores decimals, so the error is never there to compound.
Four ways to take delivery.
Self-serve queries for exploration, bulk files for pipelines, a database account for analytics stacks. Or tell us the shape and we build it.
Query API
Run SQL-style queries against the archive over HTTPS; results come back as JSON or CSV. The fastest path from question to dataframe.
Self-serve · full archive on Pro
Bulk exports
Per-dataset, per-date-range exports delivered to your S3/GCS bucket or as a direct download. One-time fee per export.
Pro quota · à la carte beyond
Read-only DB access
A scoped, read-only database account against the archive. Point your own analytics stack at it, no export step at all.
Enterprise plans
Custom datasets
Tell us the markets, date range, and shape; we build it and deliver it to your bucket. Includes backfills for ranges older than the continuous archive.
Quoted per request
OHLCV from raw trades, one curl.
The query API speaks SQL against the same one-row-per-event tables the archive stores, so candles, spreads, and PnL aggregates are a GROUP BY away.
Get an API key
Self-serve from the console. Pro opens a 30-day query window; the full archive opens on Ultra.
POST SQL to the query endpoint
data.hyperliquidrpc.com/query takes your query with the key in x-api-key. No driver or client library required.
Get candles back as JSON or CSV
This example builds 1-minute BTC OHLCV from the last hour of raw fills, the same rows the live stream delivers.
Runnable as-is. Swap in your key.
curl https://data.hyperliquidrpc.com/query -H 'x-api-key: YOUR_KEY' \ -d "SELECT toStartOfMinute(block_time) m, argMin(px,block_time) o, max(px) h, min(px) l, argMax(px,block_time) c, sum(sz) vol FROM trades WHERE coin='BTC' AND block_time > now()-3600 GROUP BY m ORDER BY m"Complete since launch. Compounding daily.
The honest framing: we lead with completeness and daily volume, not years-of-history. The archive is continuous from launch with no TTL, and it grows by hundreds of millions of rows a day.
0.00B+
order events archived
0M+
order updates / day
0.0M+
trades / day
0+
markets covered
Plus 1B+ book updates and 63,000+ distinct wallets observed, as of 2026-06-09. Need ranges older than the continuous archive? Custom backfills are available on request, quoted per markets, range, and shape.
Or skip the ETL entirely.
The aggregations most teams build from the raw archive. Already built, from the same exact tape.
Historical data, answered.
Coverage, formats, schema parity, and how the archive relates to the live feeds.
A data and RPC provider for Hyperliquid. We stream real-time market data over gRPC, serve HyperEVM/HyperCore RPC, and sell a full historical dataset for backtesting and analytics.
All of them: perpetuals, spot, and HIP-3 / pre-market, including new listings. Currently 462 markets and growing.
Exact. Prices, sizes, PnL, and fees are byte-exact decimals, never floats. It reconciles to the chain and to Hyperliquid's public API.
Yes. We keep a complete archive of every event with exact precision. Query it via API, get bulk exports (Parquet/CSV/JSONL), or get scoped read-only database access. Custom backfills on request.
Yes. Start any stream from a past block height to backfill, then continue live. Replay is available on Pro and Enterprise.
The archive is continuous since launch and grows by hundreds of millions of rows a day: 1.65B+ order events and 1B+ book updates so far, with no TTL. For ranges before the archive started, we build custom backfills on request.
Parquet, CSV, and JSONL, delivered to your S3/GCS bucket or as a direct download. The query API returns JSON or CSV. Enterprise can also get scoped read-only database access for its own analytics.
Yes. Same fields, same exact-decimal encoding. Backtest on the archive and deploy on the live feed with zero remapping.
From the full tape, not samples: OHLCV candles (1s/1m/1h/1d) from every fill, wallet PnL from attributed fills, liquidation history with victim, size, and mark price, and funding-rate time series per market.
Run the first query tonight.
Pro opens a 30-day query window; Ultra opens the full archive with bulk exports.