# AIFIRST News

> Automated, categorized, searchable AI news for humans and software agents.
> Operated by Halcer d.o.o., Ljubljana. Content is aggregated from public
> sources (Hacker News, Reddit, GitHub trending) by an AI pipeline; summaries
> and importance scores are machine-generated. Verify against the linked
> originals before relying on any item.

## Machine-readable feeds

- /data/items.json : the full news feed. JSON array of items with fields:
  id (string), title (string), url (string), source (string: hn,
  reddit-<sub>, github-trending), category (string: claude-code, models,
  tools, open-source, image-video, audio, agents, apps, research),
  importance (integer 1-10, editorial ranking; 8+ counts as breaking),
  summary (string), github (string URL or null), ts (ISO 8601 UTC).
  Items may carry note: "sample" when the live pipeline has not published
  yet; treat those as placeholders, not news.
- /data/status.json : nightly server-side status of major AI services
  (Anthropic, OpenAI, Hugging Face, GitHub). Fields: generated (ISO 8601),
  services[] with id, name, status (up | degraded | down | unknown),
  checked, detail. "unknown" means not probed, never a guess.
- /sitemap.xml : page inventory.

## Pages

- / : lander with breaking lane, category sections and client-side search
- /status : AI service status monitor
- /my-activity : account-less starred-items queue (browser-local token)
- /privacy and /terms : legal pages

## Coming (MVP-2)

Star-to-agent pipeline: starring an item queues automatic scrape, knowledge
extract, independent benchmark, pro-and-contra review and an adoption
verdict, plus an MCP-style package endpoint so agents can pull a full "file
package" per news item. Not live yet; the current queue status is only
"queued".