Google News — Localised News Search That Cites Itself

May 18, 2026

Ask Alfrada for the news the way you'd ask a research assistant — by topic, region, or recency — and it pulls articles from your country's own Google News edition, attaches thumbnails, and saves citations into your session so Presentation, Report Generator, and the rest of the toolchain can drop them in without re-searching. A query from Berlin returns the German edition. A query from London returns the UK edition. Karachi gets Pakistan-localised results. Zero parameters required.

What you can do

• Search Google News by query — google_news runs the search via SearchAPI and returns ranked articles with title, source, date, snippet, link, and thumbnail.
• Auto-localised results. Country and language flow from your detected location — gl=de and hl=de for Berlin, gl=gb for London, gl=pk for Karachi, and so on. Override with the location field when you want a different regional bias (e.g. "Tokyo, Japan").
• Top results auto-save as citations. Default 5 articles persist as .news.json files in the session out/ folder (set save_count to a different number, or 0 to skip saves). Each citation carries title, source, date, link, snippet, gl/hl context, and a ready-to-render credit line. Indexed for report_generator and presentation to consume in the same turn.
• Article thumbnails render inline. SearchAPI returns thumbnails as embedded base64 data — Alfrada decodes them once, writes the bytes as real image files in out/, and substitutes short paths in the response. Thumbnails render in the chat card without inflating LLM token cost.
• Filter by recency — last_hour, last_day, last_week, last_month, last_year. Useful for breaking-news coverage or to drop stale content from a deep-research run.
• Sort by relevance or chronology — default returns the most relevant matches; pass sort_by="most_recent" for chronological order.
• Pagination — page flips through pages of results. The card header shows the current page.
• Per-row inline citation. Every article row shows the publisher's hostname (clickable) plus, for saved articles, the citation filename — so the link between "what I read" and "what got saved" is visible without scrolling to a footer.

Where this shows up

• Briefing on a fast-moving story. "Pull the last 24 hours of coverage on the EU AI Act and give me a one-page brief with citations." Auto-filters to time_period=last_day, saves the top 5 articles as citation JSONs, then summarises with inline source links.
• Localised market scan. "What's the German press saying about Bundesliga transfer rules this week?" German edition with German-language results — not the US sports-news lens Google would default to from a US-hosted IP.
• Press coverage for a pitch. "Find 8 recent stories that mention Strategize Labs and cite each with publisher + date." Saved citations carry full attribution; report_generator picks them up in the same session.
• Pasted URL hand-off. Drop https://news.google.com/foryou into chat — the tool auto-activates and the agent can search from where you were browsing.

Try it

• "latest news about climate policy"
• "top headlines in Germany today"
• "recent coverage of SpaceX Starship launches from the last day"
• "news about EU AI Act enforcement — save the top 8 as citations"
• "breaking news on Bundesliga transfer deadline"
• "news from Pakistan about IMF programme talks, last week, sorted by most recent"

Heads up

• google_domain is deprecated by SearchAPI as of April 15, 2025. Country edition routing now flows from gl + hl exclusively — Alfrada uses the shared google_location_utils.apply_geo_params() helper, the same path Google Images and the rest of the Google-family tools take.
• No cr country-restrict. Soft geo bias (gl + hl) only — matches the Google Images choice. Adding a hard country-origin filter would starve result pools in smaller markets. If you specifically need only in-country publishers, ask the agent and it can filter the result list client-side.
• Citations are metadata, not full article bodies. Each saved .news.json carries title, source, date, link, and snippet — not the full article text. If you need the body, the agent chains to the browser tool to read the article inline (observed naturally in testing). This keeps citation save fast and avoids the ~30-50% failure rate of server-side scraping against paywalled / anti-bot publishers.
• Thumbnail rendering writes small image files to your session out/ folder. They count against the same lifecycle as every other tool's saved assets — nothing new to clean up.
• Pairs with the rest of the toolchain. Saved citations are usable by presentation (slide footnotes), report_generator (cited reports), and the browser tool (drill into one specific article).