Hyperliquid Data Source

Hyperliquid is the source of truth for the tradable asset universe and the main asset-level market context.

The backend stores all supported Hyperliquid assets, including perps and spot-style markets. Generic HIP-3 category exposure is intentionally out of scope for the product picker.

What We Store

Channel	Table	Origin	Cadence / granularity	Purpose
Mid prices	hyperliquid_mid_prices	Source	allMids stream bucketed at 5s	Short-horizon price context
OHLC candles	hyperliquid_candles	Source	candle WebSocket at 1m	Real OHLC rows for chart/context reads
OHLC rollups	hyperliquid_candles	Derived	1h from 1m candles	Larger windows without extra source subscriptions
Liquidity profile	hyperliquid_liquidity_profiles	Source-derived from L2 snapshots	5m	Spread, depth, and imbalance
Funding pressure	hyperliquid_funding_pressure	Source-derived from perp context	5m	Funding, open interest, crowding
Volatility profile	hyperliquid_volatility_profiles	Derived	5m	Realized volatility and jumpiness
Positioning pressure	hyperliquid_positioning_pressure	Derived	5m	Composite direction/crowding signal

Price Policy

Use two price paths:

allMids stream
  -> 5-second midpoint buckets
  -> hyperliquid_mid_prices

1m candle WebSocket subscriptions
  -> real OHLC rows
  -> hyperliquid_candles
  -> local 1h rollups

This avoids the REST candleSnapshot rate-limit problem for the full asset universe. Larger chart intervals should be derived locally from 1-minute candles instead of subscribing to every interval.

How We Fetch Data

Configured endpoints:

Config	Default
HYPERLIQUID_WS_URL	wss://api.hyperliquid.xyz/ws
HYPERLIQUID_INFO_URL	https://api.hyperliquid.xyz/info

Streaming price data uses one WebSocket connection:

{
  "method": "subscribe",
  "subscription": { "type": "allMids" }
}

That message returns all midpoint prices in one stream payload. We parse channel: "allMids" messages and store compact 5-second buckets in hyperliquid_mid_prices.

For OHLC, we subscribe to 1-minute candle streams per active supported symbol:

{
  "method": "subscribe",
  "subscription": {
    "type": "candle",
    "coin": "BTC",
    "interval": "1m"
  }
}

We parse channel: "candle" messages and store 1-minute rows in hyperliquid_candles. Larger intervals are derived locally from those rows.

Registry and feature inputs use HTTP POST to HYPERLIQUID_INFO_URL:

Request body	Used for
{ "type": "meta" }	Perp universe and symbols
{ "type": "spotMeta" }	Spot-style universe, display names, search terms
{ "type": "metaAndAssetCtxs" }	Perp contexts, active status, open interest inputs
{ "type": "predictedFundings" }	Predicted funding rows
{ "type": "l2Book", "coin": "BTC" }	L2 book snapshot for liquidity profile

The client still has a candleSnapshot helper for probes/backfills, but the runtime price ingestor should not use it for the full universe because it becomes request-heavy.

Current hot retention:

Data	Retention
5s mid-price buckets	30 minutes
1m OHLC source candles	24 hours
1h OHLC rollups	7 days
5m feature profiles	7 days

Asset Registry

hyperliquid_assets stores the supported-market registry:

• Exchange symbol.
• Display name and search terms.
• Asset kind: perp, spot, or unknown.
• Category/subcategory metadata for the picker.
• Coverage flags for prices, candles, profiles, and trading.

Runtime ingestion refreshes the registry from Hyperliquid metadata and stream-discovered symbols. Product selection should read this registry instead of inferring support from observation rows.

Trading Context

Trading prompts can request Hyperliquid context per enabled asset:

• Recent mids for the freshest short-horizon move.
• OHLC candles for chart structure.
• Liquidity profile for tradability and slippage risk.
• Funding pressure for crowded perp positioning.
• Volatility profile for regime and sizing.
• Positioning pressure for composite market state.

Local Commands

cd services/market-backend
npm run hyperliquid:assets
npm run hyperliquid:candles -- --duration-ms=70000
npm run hyperliquid:features
npm run hyperliquid:maintain