0xArchive

What it is

A market-data API for two perpetual DEXes and nothing else. On Hyperliquid it covers core perps, Spot, HIP-3 builder perps and HIP-4 outcome markets as separate route families; on Lighter it covers mainnet and the second deployment on Robinhood Chain. Live data arrives over a WebSocket, history over REST or over a WebSocket replay that re-emits stored events in order, and bulk ranges as Parquet files bought by the gigabyte.

The reason to look at it instead of a cross-venue archive is depth on those two venues. On Hyperliquid it serves an L4 book — every resting order with the wallet that placed it, as checkpoints plus ordered diffs — next to the usual price-level L2. On Lighter it serves an L3 book of individual orders, keyed by Lighter's own order and account indexes rather than wallet addresses. Beyond the book, it reaches into territory market-data vendors usually leave to on-chain tools: order history, take-profit and stop-loss orders, TWAP events, account positions by wallet address, and webhook alerts on fills by a wallet you name.

The operator is Archive Labs Inc., a Delaware corporation, which the about page says was founded in 2025 and is based in the United States. The vendor's own account of its origin is that Hyperliquid's history sat in public S3 buckets but took real engineering to use. The terms state that it is not affiliated with, sponsored by or endorsed by either venue.

Pricing

Four self-serve plans and an Enterprise contract. Free is $0 with no card; Build is $49 a month, Pro $199 and Scale $799, each about 20% less on annual billing. Every plan, Free included, can call every route and schema. What the plans change is capacity — credits, request rate, concurrency, WebSocket subscriptions and replay speed — and Free's 30-day history window. Build is the first plan with the full archive and the first that delivers webhooks.

The unit is a credit, and it is not a request. REST charges by rows returned with a one-credit minimum: trades, fills, L2 history, funding, open interest, liquidations and Lighter L3 at 1,000 rows per credit, L4 at 5,000, candles at 10,000, and single-value routes such as a current book snapshot at one credit flat. The WebSocket charges up to one credit per delivered message, live or replayed, and that is the number to plan around. The Lighter book channel sends up to one top-20 book a second by default — up to 3,600 credits an hour for one symbol — so a single default-rate subscription can use up Free's 50,000 monthly credits in under 14 hours. Build's 80 million covers roughly 30 such streams for a whole month.

Parquet exports are priced per gigabyte and need no subscription: $3 for L2 and L3 order books ($5 minimum), $4 for L4 books, L4 orders, trades and liquidations, and $0.50 for funding and open interest. Orders over $250 get 10% off and over $2,500 20%. Paid plans add monthly export credits — $50, $300 and $1,500 — which roll over up to twice the allowance.

Card subscriptions go through Stripe and renew until cancelled. The alternative is USDC on Base through x402, which buys 30 days and does not renew. All fees are non-refundable, and the terms allow price changes on 30 days' notice.

Data & coverage

First served dates by market type, from the venue-coverage documentation and matched against the live coverage endpoint on 7 October 2026:

• Hyperliquid core — native L2 book 15 April 2023, trades the same day, funding and open interest 20 May 2023, candles 22 March 2025, liquidations 22 December 2025, full-depth L2 and L4 11 March 2026.
• Hyperliquid Spot — trades and candles 22 March 2025, native L2 and L4 May 2026 for most pairs.
• HIP-3 builder perps — trades 13 October 2025, native L2, funding and open interest 16 February 2026.
• HIP-4 outcome markets — everything from 2 May 2026, with prices as implied probabilities.
• Lighter mainnet — trades 17 January 2025, funding and open interest 25 August 2025, L2 book 29 January 2026, L3 book 5 March 2026, liquidations 10 June 2026.
• Lighter on Robinhood Chain — trades 26 June 2026, order book 22 August 2026.

For Hyperliquid that is about eighteen months deeper than the cross-venue archives here: Tardis.dev's Hyperliquid feed starts 29 October 2024 and CoinAPI's Hyperliquid trades 27 September 2024.

The 2023 start has limits. It is the start of the native L2 book, which on Hyperliquid core and HIP-3 stops at 20 levels per side whatever depth you ask for; full depth exists only from 11 March 2026, when the L4 data begins. That native book comes from two interleaved sources, labelled on every snapshot: venue, Hyperliquid's own published book, about every 5 seconds since 20 June 2026 and faster before; and node, rebuilt from 0xArchive's own Hyperliquid node on a fixed grid, every 500 ms since 25 July 2026 and every second before that. Lighter's L3 book is capped at 250 orders per side per tick.

Completeness is something you query rather than assume. Per-symbol coverage routes report windows, gaps and observed cadence by data type; an incidents route lists past interruptions and the symbols they touched; and freshness routes report each data type's lag behind now. The coverage summary read on 7 October 2026 listed 234 Hyperliquid core symbols with order books at about 1.2-second resolution, and data current to the minute.

Lighter mainnet is not only crypto. Its markets include equities and indices, FX, commodities and pre-IPO markets, and the Robinhood Chain deployment quotes tokenised stocks such as AAPL-USDG against USDG.

Integrations

First-party clients in three languages plus a CLI, all named in the vendor's own reference: oxarchive on PyPI, @0xarchive/sdk on npm, the oxarchive crate for Rust, and @0xarchive/cli, whose command is oxa. The source sits under the 0xArchiveIO organisation on GitHub with MIT licences, and the PyPI package lists the vendor's support address as its author, so the obvious names are the vendor's own. The Python client has had 49 releases, from 0.1.0 on 11 January 2026 to 1.12.0 on 5 October 2026.

The MCP server is hosted at mcp.0xarchive.io over Streamable HTTP, and it serves market data rather than documentation: tools for discovering and retrieving data under the mcp:market.read scope, plus optional scopes to read and write webhook rules. Sign-in is OAuth, with no API key. An unauthenticated POST to the endpoint returned 401 on 7 October 2026 and a made-up path on the same host returned 404. The older self-hosted stdio package on GitHub is marked retired.

Webhook alerts send a signed POST, or a Slack or Discord message card, on liquidations, funding flips, open-interest moves, HIP-4 settlements, new listings, activity by watched wallets and finished exports. They start at Build — one endpoint, eight rules and 5,000 deliveries a day — and deliveries retry until your receiver answers 2xx or 24 hours pass.

REST returns JSON, with Apache Arrow as an opt-in for the trades family; exports are zstd Parquet with a README describing every column. The REST contract is published as OpenAPI 3.0 and the WebSocket as AsyncAPI 3.0, and every metered response carries rate-limit and credit headers.

Limitations

Two venues. No centralised exchange, no other DEX. That is the right trade for a strategy that exists because of Hyperliquid's book, and the wrong one for anything comparing venues — a basis trade against Binance needs a second vendor.

Free is a 30-day window, and streaming burns credits fast. Free is a real tier for building an integration. It is not a tier for running a live feed: see the credit arithmetic under Pricing.

"Since April 2023" covers part of the dataset. It is true of native top-20 L2 and of trades over REST. Candles start in March 2025, full depth and L4 in March 2026, and Hyperliquid trades bought as files start in March 2025, because the vendor will not sell the earlier fill history it calls incomplete. Check coverage_by_type for the symbol and data type before planning a backtest around a date.

Redistribution is barred, and most other uses are not. You may not resell raw records, publish bulk dumps, mirror the API or run a substitute feed, and benchmarks, indices and financial products need written permission. Model training and customer-facing products are explicitly permitted when the output is materially transformed — looser than most archives in this category — and data downloaded while entitled stays usable after cancellation.

A young operator with no named people. The company dates from 2025 and the GitHub organisation from January 2026. Neither the about page nor the terms name a founder or an employee, the mailing address is a PO box, and 72 of 75 commits on the Python SDK come from one pseudonymous account. There is outside use: Arrakis Research credits 0xArchive as a data source in two research posts on its own blog. The terms warrant nothing about completeness and cap liability at the fees paid in the previous twelve months.

Export file names omit the venue. BTC trades from Hyperliquid and from Lighter in one order arrive under the same file name, so save them to separate folders and read the venue from the file metadata.

Alternatives

Tardis.dev is the archive to choose when the strategy spans venues: 64 feeds of raw exchange messages, Hyperliquid among them from October 2024. It sells no hosted real-time feed, its plans start at $350 a month, and its licence limits model training to internal, task-specific models.

CoinAPI puts Hyperliquid next to 400-plus venues under one symbol scheme, with L4 order-lifecycle events over its streaming endpoint, but has no free plan and starts its Hyperliquid trades in September 2024.

Laevitas also derives a Hyperliquid surface from node data — fills, liquidations, funding, TWAP events, resting orders and wallet positions — at $500 a month for a key or 0.10 USDC a call over x402, and adds options Greeks and a volatility surface, which this card does not touch.

If current state is enough and history is not the job, the Hyperliquid Python SDK reads the venue's own API for nothing; the how-to on Hyperliquid positions and trade history shows where that free history runs out. The rest of the market data APIs listing covers the cross-venue providers.