Hyperliquid API for AI Agents vs Arrays: What Circle's Agent Can't See

What Circle shipped on 1 September, and why Arrays is now the default

On 1 September 2026 Circle published Build a Trading Thesis Agent with Arrays and Circle Agent Stack. It is the first end-to-end tutorial for the Circle Agent Stack: an Agent Wallet that holds USDC under spending policies, an Agent Marketplace the agent searches at runtime, and Nanopayments that settle sub-cent USDC calls gas-free on Base. Crypto Economy covered it the next day. Etherscan's MCP server shipped the week before. Binance Agent OS landed on 20 August. Everyone is teaching agents where to fetch.

The tutorial's data provider is Arrays, which lists 15 USDC-payable services on the marketplace. The prompt tells the agent to search the marketplace for Arrays crypto market data services, inspect the price, pay per call, and write a thesis. That makes Arrays the default paid feed for anyone who copies the prompt, and a lot of people will: the six starter kits (langchain, claude-agent-sdk, openai-agents, vercel-ai, google-adk, mastra) ship no data adapter at all. The agent discovers providers by calling circle services search, so whatever the tutorial names is what gets bought.

This is a good tutorial. The research-only framing is right, the receipt log is right, and pay-per-call is the correct model for a one-off pull. The question for a Hyperliquid API for AI agents vs Arrays is narrower: what does that agent actually see once it has paid?

What the Circle plus Arrays stack sees

Circle's research plan pulls nine things. Here is the list, verbatim in order, with what Arrays' own docs say the feed covers.

• Token detail: metadata profile by symbol.
• Fear and Greed Index: single daily value.
• Funding rates: Binance perps, plus a separate Hyperliquid funding endpoint (hourly, funding_rate and timestamp only).
• Open interest: documented as Binance-only.
• Long/short ratio: documented as Binance-only.
• Taker buy/sell volume: Binance.
• Price and volume: Binance spot and perp klines, plus Hyperliquid perp and spot candles.
• Unlock events: token unlock schedule.
• Compare and write the thesis.

So, to be exact: Arrays does have Hyperliquid data. It has HL candles and HL funding. What it has is the same two series every free SDK exposes. Its open-interest, long/short and exchange-flow feeds are Binance. Its only whale metric is a BTC-only on-chain Exchange Whale Ratio, an inflow ratio, not a classified account. And its pay-per-call catalogue is crypto-only today; Arrays' docs list ETF, stocks, macro and options as planned for the next two quarters on that path, so a Circle agent cannot buy ETF flows from Arrays at all yet.

That is a candle-and-funding view of Hyperliquid. It is the axis on which every provider looks the same, and it is the one axis we deliberately do not compete on.

What does a Hyperliquid API for AI agents need that Arrays cannot sell?

Everything below was checked on 7 September 2026 against Arrays' public docs index and against our production API with a real key (every row returned 200). Rows are grouped by whether Arrays sells the signal.

The four bold rows are the ones a Hyperliquid thesis lives or dies on. On the day we ran the check, /quant/whales reported net_bias: "short" across 3,570 classified accounts with total_net_usd around -$551M, while /regimes/current said btc_led_bull and /quant/market said squeeze at 0.975 confidence. Whales net short into a confirmed uptrend that the HMM calls a squeeze is exactly the divergence a thesis agent is supposed to surface. A candle-and-funding stack cannot see any of it.

The research-only prompt: same skeleton as Circle's, plus the tape

This mirrors Circle's Config / Goal / Setup / Research plan / Rules / Return layout on purpose, so an agent that ran their prompt can run this one unchanged. It places no orders, holds no wallet, and never contains a key. Step 1 needs no key at all. It also lives in the prompt library and the GitHub mirror.

Use CryptoDataAPI to build a research-first trading thesis, then show what the same thesis would have missed if it had only used Arrays via the Circle Agent Marketplace.

Config:
- Asset symbol: <SYMBOL>
- Time window: <TIME_WINDOW>
- Output: markdown

Goal:
Collect the same signal set Circle's 1 Sep 2026 thesis-agent tutorial pulls from Arrays, then add the Hyperliquid tape and quant signals Arrays does not sell. Return a thesis brief plus a "What Arrays missed" table. Do not place a trade. Do not give financial advice. Do not put an API key inside this prompt; keep it in the X-API-Key header or the MCP config.

Setup:
1. GET https://cryptodataapi.com/api/v1 (no key). Read the endpoint map and note which paths your key tier unlocks.
2. If you have no key: POST https://cryptodataapi.com/api/v1/auth/keys with {"email": "<EMAIL>"} (no key needed, returns a cdk_live_ key once). Confirming the email lifts the key to 1,000 requests/day and switches Pro on for 24 hours.
3. Or connect the MCP server: claude mcp add --transport http cryptodataapi https://cryptodataapi.com/mcp  (see https://cryptodataapi.com/ai-agents).
4. Send a named User-Agent. Read the X-API-Version header once and record it.

Research plan, part A (the Arrays-equivalent set, all free tier):
1. Sentiment: GET /api/v1/sentiment/fear-greed  -> value, classification
2. Funding: GET /api/v1/hyperliquid/funding-rates?symbol=<SYMBOL>&limit=24  -> current_rate, history[]
3. Open interest: GET /api/v1/hyperliquid/open-interest  -> assets[].open_interest_usd for <SYMBOL>
4. Long/short: GET /api/v1/derivatives/binance/long-short-ratio?symbol=<SYMBOL>USDT  -> long_short_ratio, top_trader_long_short_ratio
5. Taker flow: GET /api/v1/market-intelligence/taker-buy-sell  -> data[].exchange_list[].buy_ratio
6. Unlocks: GET /api/v1/supply/unlocks  -> items[] for <SYMBOL> (free = next 7 days, Pro = full horizon)
7. Events: GET /api/v1/event/calendar  -> items[].{event_type, bias, magnitude}. Page: https://cryptodataapi.com/calendar

Research plan, part B (what Arrays does not sell):
8. Classified Hyperliquid whales (Pro): GET /api/v1/quant/whales  -> summary.{accounts_tracked, total_net_usd, net_bias, long_short_ratio}, top_coins[] for <SYMBOL>. Page: https://cryptodataapi.com/quant-whales
9. Positioning by trader class (Pro): GET /api/v1/quant/positioning  -> <SYMBOL>.by_type.{whale, smart_money, market_maker, high_leverage}.net_usd
10. Perp gamma exposure and flip regime (Pro): GET /api/v1/quant/gex  -> regime, flip_line, per-coin gex for <SYMBOL>. Page: https://cryptodataapi.com/quant-gamma
11. HMM regime nowcast (Pro): GET /api/v1/quant/market  -> regime.{label, confidence}, probabilities.{directional, volatility, liquidation_risk}
12. Long-horizon regime (free): GET /api/v1/regimes/current  -> regime, rationale. Page: https://cryptodataapi.com/regimes
13. Liquidations, 24h by venue (free, BTC-scoped on free): GET /api/v1/market-intelligence/liquidations  -> long_liquidation_usd_24h, short_liquidation_usd_24h. Page: https://cryptodataapi.com/liquidations
14. Per-event Hyperliquid liquidation tape (Pro): GET /api/v1/backtesting/hl-liquidations?coin=<SYMBOL>&start=<ISO8601>&limit=200  -> data[].{time, coin, side, px, sz, usd, method, mark_px}
15. Spot ETF flows (free, BTC/ETH/SOL): GET /api/v1/market-intelligence/etf/btc/flows  -> flows[].flow_usd. Page: https://cryptodataapi.com/etf-flows

Rules:
1. Do not execute trades or generate orders.
2. Do not invent data. Cite the endpoint and field next to every number, e.g. (quant/whales summary.net_bias).
3. If a call returns 401, 403 or 429, record the status and the upgrade block in the receipt log, mark the row unavailable, and continue.
4. Treat every step in part A as a signal Arrays also sells. Treat every step in part B as a signal Arrays does not sell. Say so in the table.
5. Treat the output as research, not financial advice.

Return:
- api version (X-API-Version)
- endpoints used, with tier
- signal summary (part A)
- signal summary (part B)
- "What Arrays missed" table: columns = signal | value now | source endpoint | Arrays equivalent (yes / partial / none) | changed the thesis? (yes / no, one line why)
- thesis, bull case, bear case, invalidation condition, confidence level, caveats
- receipt log: every request, status code, bytes, and whether it was free or Pro

The forced "What Arrays missed" table is the point. It makes the agent write down, per signal, whether the Circle stack could have bought it and whether it changed the thesis. Most of the time the free rows agree with each other and the Pro rows are where the disagreement lives.

One curl to prove the first Pro row exists

Everything in part B is a plain GET. This is the whale row from the table, with the fields the prompt asks the agent to cite.

curl -s https://cryptodataapi.com/api/v1/quant/whales \
  -H "X-API-Key: cdk_live_YOUR_KEY" \
  -H "User-Agent: my-thesis-agent/1.0" \
  | python -c "import sys,json; d=json.load(sys.stdin)['summary']; print({k:d[k] for k in ('accounts_tracked','universe_size','total_net_usd','net_bias','long_short_ratio')})"
# {'accounts_tracked': 3570, 'universe_size': 14707, 'total_net_usd': -551219342.1, 'net_bias': 'short', 'long_short_ratio': 0.872}

With a free, unconfirmed key that call returns a JSON 403 whose upgrade block names the tier, the live price and the subscribe link. The agent should log it and move on; the prompt says so. Confirm the email on the key and the same call returns 200 for the next 24 hours.

Pay-per-call is right for a one-off pull. It is wrong for a polling dataset

Arrays' nanopayment rates are $0.008 per basic call and $0.03 per advanced call, in line with what metered whale and quote feeds charged over x402 through August. This is not a complaint about Arrays' pricing. It is about what a thesis agent does after the first thesis.

• A single research brief is maybe 15 calls. At marketplace rates that is under 50 cents. Pay per call.
• A thesis agent that watches re-reads whales, GEX and the regime on a clock. One dataset polled every 5 minutes is 8,640 calls a month: $69 to $86 a month for one feed at $0.008 to $0.01, or $259 at the advanced rate.
• Pro is a flat $39 a month for all of the Pro rows above at 10,000 requests a day, and the GEX, whale and regime feeds are the polling feeds.

So the honest split: let the agent pay per call for the discovery pass, then subscribe the moment it starts polling. We sell the subscription over x402 too: POST /api/v1/payments/agent-subscribe returns a 402 with the USDC amount, and a signed Base USDC payment comes back with a Pro key, gasless. Our 403 body says the same thing at the exact moment an agent hits the paywall, because that is when it is price-shopping.

How to use this if you are following the Circle tutorial

1. Keep the Circle prompt for the wallet, marketplace and receipt-log steps. They are good.
2. Paste the prompt above as the research plan, or pull it from the prompt library.
3. Step 1 is keyless. For the rest, mint a free key with POST /api/v1/auth/keys or connect the MCP server from cryptodataapi.com/ai-agents. Confirm the email and Pro is on for 24 hours, which covers every Pro row in the table.
4. Read the returned What Arrays missed table. If no Pro row changed the thesis, stay on the free tier. If one did, that is the row you subscribe for.
5. The 24-hour Pro trial reaches every Pro row in the table, including the per-event liquidation tape. Whale and regime history stay Pro Plus.

None of this is financial advice, and none of it places a trade. It is the same research-only posture Circle asked for, with more of the Hyperliquid tape in view.

Frequently asked questions

Does Arrays have Hyperliquid data?

Yes, in part. Arrays' docs list Hyperliquid funding rates and Hyperliquid perp and spot candles. Its open interest, long/short ratio and exchange-flow feeds are documented as Binance-only, and it has no classified Hyperliquid whale data, no per-event liquidation tape, no perp gamma exposure and no regime model.

Is the research-only prompt safe to run?

It places no orders, holds no wallet and contains no API key. Every step is a GET request. The prompt tells the agent to treat the output as research, not financial advice, and to log any 401, 403 or 429 rather than retry.

Do I need an API key to run it?

Step 1 (GET /api/v1) needs no key. The free-tier rows need a free key, minted with POST /api/v1/auth/keys and no signup. Confirming the email switches Pro on for 24 hours, which unlocks the whale, positioning, gamma exposure and HMM regime rows.

Which rows are Pro versus Pro Plus?

Pro: classified whales, positioning by trader class, perp gamma exposure, the HMM regime nowcast and the per-event Hyperliquid liquidation tape. Pro Plus: whale and regime history and the daily Parquet archive. Everything else in the table is free.