0xArchive comparison · Delivery
The same Hyperliquid and Lighter history reaches you four ways on 0xArchive: the REST API, WebSocket streams, Parquet file exports, and agent tools (an MCP server, a CLI, SDKs and a Claude Code skill). Pick REST to ask one question at a time, WebSocket for a live book or an in-order replay, exports for a window you will read again and again, and the agent tools when an AI assistant does the asking.
The data is the same archive behind all four. What changes is what you keep at the end: a response, a running book, a file, or an answer.
Pick the delivery by what you keep at the end
| You keep | Use | Billed as |
|---|---|---|
| A response in your app | REST API or an SDK | API credits |
| A live book, or a replay in order | WebSocket | Plan subscriptions |
| Files you read again | Parquet export | Per GB, once |
| An AI assistant's answer | MCP server, CLI or skill | Your plan |
The REST API answers one bounded question per call
One book snapshot, one page of trades, one funding series, one coverage check. Each call names a venue, a market and a UTC window, and long windows page with a cursor. It is the right path when your code asks for data while it runs.
The SDKs for Python, TypeScript and Rust wrap the same routes with typed requests, paging and retries. They do not change what you get, only how much HTTP you write.
WebSocket carries live books and replays history in order
Subscribe to a channel and the book, trades, funding or open interest arrive as they change. The same connection can replay a stored window in sequence, faster than real time (50x on the Build plan), so a strategy sees the past the way it would have seen it live.
Hyperliquid L4 replay starts from a full book checkpoint and applies every order change after it. That is how you rebuild the exact resting book at any moment since March 2026, with the order ID and wallet on each order. Lighter streams its book, trades, funding and open interest live; its L3 order-level book and candles are replay only. The channel reference lists each one.
Parquet exports fit a window you read many times
A Data Catalog export is one Parquet file per market and data type for the UTC dates you pick, with a README that documents every column. You pay a quote per GB once, download from the dashboard, and read it with pandas, Polars or DuckDB as often as you like.
| Dataset | Export price | Minimum order |
|---|---|---|
| L4 order book, L4 order events | $4 per GB | $12.50 |
| Trades, liquidations | $4 per GB | $7.50 to $12.50 |
| L2 order book, Lighter L3 | $3 per GB | $5 |
| Funding, open interest | $0.50 per GB | $2.50 |
Prices as listed on the pricing page on 24 October 2026. Exports suit a backtest you rerun with new settings, a dataset several people share, or a warehouse table. If you need a window once, the API on your plan is the simpler path.
AI assistants use the same archive through MCP, the CLI and a skill
The hosted MCP server at mcp.0xarchive.io gives Claude, ChatGPT, Codex, Cursor and other MCP clients typed tools for every venue. You sign in with OAuth, so no API key goes into the assistant.
For agents that work in a terminal, the oxa CLI returns JSON and exit codes a script can branch on, and the Claude Code skill teaches the assistant which route answers which question. Code generators can start from the OpenAPI spec.
Every path covers the same venues and start dates
Hyperliquid core, Spot, HIP-3 builder perps and HIP-4 outcome markets stay separate in requests, streams and files, and so do Lighter and Lighter on Robinhood Chain. Start dates belong to the dataset, not the delivery path.
| Dataset | Hyperliquid core | Lighter |
|---|---|---|
| Order book | From 15 April 2023, 20 levels a side | From 29 January 2026 |
| Order-level book | L4 from 10 March 2026, with wallets | L3 from 5 March 2026 |
| Trades | Complete from 22 March 2025 | From 17 January 2025 |
| Funding and open interest | From 20 May 2023 | From 25 August 2025 |
| Liquidations | From 22 December 2025 | From 10 June 2026 |
Hyperliquid trades before 22 March 2025 exist but are partial. Exact starts vary by market, and venue coverage lists each one.
Route references
Use the REST reference for routes and paging, the WebSocket reference for live channels and replay, replay and reconstruction for rebuilding a book, and export schemas for the columns in each file.
For current API availability and plan details, use the Hyperliquid data API, data catalog, pricing, and status.