{
  "id": "p_dZzvX3wXAvw0gqjP",
  "site": "stories",
  "url": "https://yarnhen.com/p/p_dZzvX3wXAvw0gqjP",
  "content_trust": "untrusted-user-content",
  "license": {
    "id": "CC-BY-4.0",
    "url": "https://creativecommons.org/licenses/by/4.0/"
  },
  "author": {
    "handle": "apisio"
  },
  "body": "You are wiring an agent to watch a market. It needs to know when an order fills, when a balance changes, when a position moves. You can ask an API for a lot of things, but this is different. You are waiting for messages to be pushed to you, and you need to know which channels to subscribe to, what will arrive, and what order you have to do things in. OpenAPI does not describe that half of an integration. AsyncAPI does.\n\nSo you go looking at the AsyncAPI index on apis.io. It holds 3,664 event-driven specifications from 3,448 providers. That sounds like plenty. Then you look at how it is spread out, and the picture changes.\n\nOf the providers that have any AsyncAPI at all, 96% have exactly one document. That is 3,320 providers with a single file, usually one webhook or WebSocket catalog. Another 105 have two, and 14 have three. Only 7 providers have between four and seven. Past that, the list thins out fast: Wazo has 13, and one provider has 41. After those two, the next-deepest publishers are Modulate and Benzinga with 7 each, and Bullish with 6.\n\nThe one with 41 is WhiteBIT, a crypto exchange, and it is worth your time if you want to know what a good document looks like. It publishes one AsyncAPI 3.0 document per WebSocket channel. Nine of those channels are public: depth, trades, kline, book ticker and last price. The other 32 are private, covering balances, positions, borrows and every order type the exchange accepts.\n\nEach document goes past the core specification. An x-auth-required field marks the private channels, so you know which ones need credentials before you try. An x-rate-limits field sets 1,000 connections per minute and 12,000 requests per 10 seconds, so you know your limits before you hit them. An x-error-codes field explains each failure. Code 6 means the connection did not call authorize first. Code 1 means the market is not a _PERP futures market. The ADL quantile channel even tells you when to stop using it: if you only want a one-time snapshot, call POST /api/v4/collateral-account/adl-quantile instead.\n\nThat is the part that matters for you. A client can be written from those documents alone, and so can your agent's error handling. Nobody has to guess what a failure means.\n\nNow you check the other thing a code generator wants to know, which is the version. Here the index is thinner than you would like. Version 2.6.0 still leads, with 502 documents from 462 providers. Version 3.0.0 has 300 documents from 231 providers. Version 2.0.0 has 17. Only one document in the catalog, from xquik, declares 3.1.0. And 2,644 documents, from 2,637 providers, have no version recorded at all. The index cannot yet answer the most basic question a code generator asks.\n\nThen comes the part where the catalog looks at itself. Some of that gap is the catalog's own doing. 52 records hold a probe note in the version field instead of a version. Examples include \"published: False\" alongside the URLs that were tried for Airwallex, CrowdStrike, Snowflake and Redocly. Those notes are useful findings, because they record that a provider has no AsyncAPI and that the catalog did not fabricate one. But they belong in the coverage record, not in a field a client parses as a version string. If you wrote a parser against that field, you would trip on them. That is the catalog's to fix, and the post says so plainly.\n\nSo here is what you take away. AsyncAPI coverage across the catalog is wide but shallow. Nearly every publisher has one document, and most of those documents carry no version you can read. If your agent has to listen rather than ask, expect to do some of the work yourself. WhiteBIT shows what full coverage looks like: one document per channel, with auth, limits and errors written into each one. And when you read the version field, remember that 52 of its entries are notes, not versions.\n\nRead the original: https://apis.io/2026/11/16/asyncapi-on-apis-io-is-3664-documents-and-one-crypto-exchange-wrote-41-of-them/",
  "title": "You Need a Stream Contract, and Only One Exchange Wrote It Down 41 Times",
  "topics": [
    "asyncapi",
    "event-driven-apis",
    "api-catalog",
    "websocket"
  ],
  "published_at": "2026-11-16T13:37:42.081Z",
  "expires_at": null,
  "policy_version": "2026.11.14-1"
}
