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.
News-feed syndication
three.ws announcements syndicate outward through the public RSS feed.
News items live in data/rss/items.json
(edited in the repo and shipped with a deploy), and
api/rss/announcements.js serves them at
https://three.ws/rss/announcements.xml. Anything that can read RSS can
mirror the feed; no per-target credentials or push integrations are
involved.
The former in-app publishing panel and its push lanes (Dev.to, Medium,
WebSub pings) were removed along with the admin surface. Historical
items in data/rss/items.json may still carry a syndication field
recording where they were mirrored at publish time; it is inert
metadata.
Feed URLs and sources
https://three.ws/rss/announcements.xml and the shorter
https://three.ws/rss.xml serve the same feed. Both accept one optional
query parameter, source, which selects what the feed is built from:
?source= |
Feed contents |
|---|---|
omitted, or curated |
The hand-edited items in data/rss/items.json. This is what HackerNoon and every public subscriber reads. |
trythreews |
Announcements archived from @trythreews. |
nichxbt |
Announcements archived from the @nichxbt account. The account is suspended (reported 2026-08-27) and the source is unaffected: the items are read from the archived JSON in data/x-archive/, never scraped live. Only the per-item links into X are dead. See x-accounts.md. |
archive |
Both archived accounts, merged and de-duplicated, newest first. |
The value is case-insensitive. Anything else is a 400 with
{"error":"unknown_source"} rather than a silent fall-back to the
curated feed, so a subscriber who mistypes a source finds out instead of
quietly receiving a different feed forever. Responses are cached for ten
minutes at the edge and are application/rss+xml; the endpoint is
read-only and answers GET, HEAD, and OPTIONS only.
Fetch it from anywhere:
curl -s 'https://three.ws/rss/announcements.xml' | xmllint --noout - && echo 'valid feed'
Why feed validity is load-bearing
XML has no partial failure: one illegal byte anywhere in the document
makes the whole feed unreadable, so a single bad post takes every
subscriber offline rather than dropping one item. The builder in
api/_lib/rss-feed.js therefore neutralizes
the three inputs that can do that, and
tests/rss-announcements.test.js
pins each one:
- A
]]>inside an item body, which would close theCDATAsection early. It is re-emitted across two sections so it survives as text. - Characters XML 1.0 forbids (control bytes other than tab, newline and carriage return; unpaired surrogates). Escaping cannot rescue these, so they are dropped. Real astral characters and emoji are kept.
- Prose that spells the linkifier's internal URL placeholder. The placeholder is a form escaped text cannot contain, so a post can never be mistaken for a link the builder set aside.
Editing data/rss/items.json by hand is safe against all three, but run
the curl | xmllint check above after a deploy that touches the feed.
HackerNoon
HackerNoon is one of the world's largest independent tech publications.
Their importer auto-pulls from https://three.ws/rss/announcements.xml,
so any item added to the feed is picked up on their next crawl with a
canonical URL pointing back to three.ws/news/<slug>. Nothing to
configure on our side beyond keeping the feed valid.
See also: Partners, Listings and distribution.