L2 Order Book

This is the default book surface for pricing: market-making quote engines, spread monitors, execution algos that need depth, and any dashboard that shows a ladder. You choose the depth (n_levels, 1–100), the server does the aggregation, and every message is a consistent post-block view that is guaranteed never crossed. There are no “bid above ask” states to defend against.

If you need to know whose orders sit at each level, or your model depends on queue position, use the L4 book instead. If you only need a single price, n_levels: 1 here is a perfectly good top-of-book ticker.

// stream.hyperliquidrpc.com:443 — gRPC over TLS (HTTP/2)// auth: API key in x-api-key metadata · server reflection enabledpackage hyperliquid; service OrderBookStreaming {  // Full snapshot first, then one full refresh per block. Never crossed.  rpc StreamL2Book (L2BookRequest) returns (stream L2BookUpdate);}

This service is wire-compatible with QuickNode's order-book proto, so generated stubs work unchanged after swapping the endpoint and key.

# Streams JSON-encoded messages until you disconnect.grpcurl -H 'x-api-key: YOUR_KEY' \  -d '{"coin":"BTC","n_levels":20}' \  stream.hyperliquidrpc.com:443 \  hyperliquid.OrderBookStreaming/StreamL2Book

coin is required and takes exactly one market per subscription; open one stream per coin you track. Bids and asks arrive best-price-first in levels.

Fields typed string that carry numbers are byte-exact decimal strings, verbatim from the node. Compare and store them as strings or arbitrary-precision decimals. Never parseFloat a price you intend to reconcile.

• coin: required, one market per subscription
• n_levels: 1 to 100

Filters are applied on the server, so filtered-out messages never leave our edge. You only pay for what you subscribe to. Request shapes for every stream family are on Filtering; backfill semantics on Replay & backfill.

Snapshot-then-updates: first message is a full book snapshot, then one full refresh per block. The book is guaranteed never crossed.

• Each message is a full refresh, not a diff. Replace your local book state wholesale on every update. There is nothing to merge, which also means a dropped message costs you one block, not book integrity.
• One coin per subscription. The request takes a single coin. Multiplex by opening one stream per market; updates are one-per-block, so even dozens of subscriptions stay cheap.
• Cadence is the block, not the order. You see the book as of each block, not every intra-block change. For order-level granularity use the L4 book or raw book diffs.
• Never crossed is a guarantee. Best bid is always strictly below best ask. You can drop the defensive checks other feeds force on you.

{  "coin": "BTC",  "block_height": 612408133,  "time": "2026-06-09T08:14:02.418Z",  "is_snapshot": false,  "levels": [    [      { "px": "67011.0", "sz": "3.41280", "n": 14 },      { "px": "67010.0", "sz": "1.20019", "n": 6 }    ],    [      { "px": "67012.0", "sz": "2.88457", "n": 11 },      { "px": "67013.0", "sz": "5.10022", "n": 19 }    ]  ]}

One update per block per subscribed coin gives predictable, low-bandwidth depth. Use n_levels: 1 as a top-of-book ticker.

• L4 order bookEvery resting order with its owner, for queue position and microstructure.
• Raw book diffsEvery book delta for exact reconstruction at any block.
• FilteringRequest shapes and filter options for every stream family.
• BTC price tickerAn even cheaper single-number feed when depth is overkill.