Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

docs(marketing): draft the X copy for the crypto news archive with the live numbers pinned

nirholasa42868859be2
diff --git a/docs/news-archive-x-post.md b/docs/news-archive-x-post.md
new file mode 100644
index 000000000..92c83fc58
--- /dev/null
+++ b/docs/news-archive-x-post.md
@@ -0,0 +1,87 @@
+# X post: the three.ws crypto news archive
+
+Copy for announcing that three.ws runs the largest open crypto-news archive and feeds it to the
+3D agents as market memory.
+
+**Verified numbers (pulled live from `GET /api/news/archive?stats=true` on 2026-08-16).** Re-pull
+before posting; the hourly archiver appends and the count only goes up.
+
+| Claim | Live value | Source |
+|---|---|---|
+| Archived articles | 740,889 | `stats.total_articles` |
+| First article | 2017-09-23 (not 2016) | `stats.first_article_date` |
+| Last article | today, refreshed hourly | `stats.last_article_date`, [api/cron/news-archive-append.js](../api/cron/news-archive-append.js) |
+| Months covered | 103 | `months.count` |
+| Languages | English + 中文 | `stats.languages` |
+| Live publisher feeds | 197 | [api/_lib/news-sources.js](../api/_lib/news-sources.js) `NEWS_SOURCES` |
+| Agent memory path | every read story lands in the durable `news_knowledge` corpus that grounds the 3D agents | [api/_lib/news-knowledge-store.js](../api/_lib/news-knowledge-store.js), `GET /api/news/knowledge` |
+
+**Do not say "since 2016."** The corpus starts September 2017. Do not say "every article ever
+published" either: 197 feeds is the live registry, and the archive is enriched (tickers,
+sentiment, entities, market context at publish time), which is the actual differentiator.
+
+Links that resolve: three.ws/markets/archive, three.ws/markets/news, three.ws/markets/digest.
+
+---
+
+## 1. Main post
+
+> Did you know three.ws runs the largest open crypto news archive?
+>
+> 740,000+ articles from 197 crypto publishers, September 2017 to this hour, English and Chinese,
+> every one enriched with tickers, sentiment, entities, and the BTC/ETH price at the moment it
+> was published.
+>
+> We built it because our 3D AI agents needed memory. An agent that only sees today's headline
+> has no idea it is the fourth time that narrative has run. Ours can look up every prior cycle,
+> what was said, who said it, and what the market actually did next.
+>
+> Free to search: three.ws/markets/archive
+
+---
+
+## 2. Shorter variant (single-screen)
+
+> Most AI agents read today's news. Ours read nine years of it.
+>
+> three.ws keeps the largest open crypto news archive: 740k+ articles, 197 publishers, Sept 2017
+> to this hour, EN + 中文, each tagged with tickers, sentiment, and the market price at publish
+> time. It is the memory layer behind our 3D agents.
+>
+> three.ws/markets/archive
+
+---
+
+## 3. Thread (4 posts)
+
+> **1/** Did you know three.ws has the largest open crypto news archive?
+>
+> 740,000+ articles. 197 publishers. September 2017 to this hour. English and Chinese.
+>
+> It is not a scrape sitting in a bucket. Here is what it is for.
+
+> **2/** Every article is enriched on the way in: tickers mentioned, sentiment, entities, and the
+> BTC/ETH price and fear/greed reading at the exact moment it published.
+>
+> So you can ask what the market was actually doing when a story ran, not just that it ran.
+
+> **3/** That corpus is the memory for our 3D AI agents.
+>
+> An agent with only today's feed treats every narrative as new. An agent with nine years of
+> coverage knows this exact story ran in 2018, 2021, and 2024, and what followed each time.
+> Same model, better context.
+
+> **4/** All of it is searchable by keyword, ticker, source, sentiment, date, and language, and
+> queryable through our API for your own agents.
+>
+> three.ws/markets/archive
+
+---
+
+## 4. Reply to have ready
+
+If someone asks how it stays current: the archiver runs hourly against all 197 feeds, so the
+newest month is always live and the corpus grows every hour. If someone asks about publisher
+rights: the reader shows a bounded excerpt plus our own summary and sends readers to the
+publisher, and the cap plus any takedown is enforced server side across the reader, the RSS
+mirror, archive search, and the agent knowledge corpus. See [docs/news-rights.md](news-rights.md).