Local News API changelog

2026-11-11

Content type filter

Content type: filter articles by the kind of page they are with the new content_type parameter on Search, Search advanced, Latest headlines, and Latest headlines advanced. Pass one or more of news_report, analysis, interview, opinion, press_release, listicle, service, commerce, review, autogen, seo_spam, obituary, ugc, and non_article, as a comma-separated string or an array. news is a shortcut for news_report, analysis, and interview, and all applies no filter. See Content types.
Response: every article includes its label in nlp.content_type.
Availability: labels are available for articles from October 17, 2026. Later in Q4 2026, articles back to January 1, 2026, will be labelled too. Articles without a label are left out when the filter is set. Requests without content_type return articles of every content type, as before.

2026-10-23

Countries: filter articles by the country where the publisher is located with the countries parameter on the search and latest headlines endpoints. Pass two-letter ISO 3166-1 alpha-2 codes as a comma-separated string ("US,CA") or an array. Codes are case-insensitive. When sources is also set, countries is ignored.
Fixed: wildcard terms with accents now match on language-specific search fields. For example, q: "éolien*" with lang: "fr" now finds éolienne. Both the accented and unaccented forms are searched, so results can only grow.
Fixed: a date range that starts more than 30 days ago and ends within the last 30 days now returns articles from the whole range. Before, articles older than 30 days were left out when to_ was recent.

2026-10-08

Fixed: geonames[].detection_methods on the advanced endpoints now accepts the documented names: dedicated_source, local_section, regional_source, standard_format, proximity_mention, and ai_extracted. Before, filtering with these names matched nothing. See GeoNames filtering.
Fixed: responses return geonames[].detection_methods with the same documented names, without duplicates.
Changed: an unknown value in geonames[].detection_methods now returns a validation error instead of matching nothing. Values that worked before still work.

2026-02-25

Clustering: articles published on or after January 1, 2026, are clustered with a new method based on graph clustering of Qwen embeddings. Date ranges that end before January 1, 2026, use the previous method. The start of the range (from_ or when) decides which method runs.
Clustering threshold: set the similarity threshold for clustering with the clustering_threshold parameter on the advanced endpoints. It must be greater than 0 and at most 1. The default is 0.7 for ranges from January 1, 2026, onward and 0.6 for older ranges.
Embeddings: qwen_embedding is now accepted in embeddings_output on the search, latest headlines, and search by endpoints, for plans that include embeddings.
Breaking: a request with clustering: true whose date range crosses January 1, 2026, 00:00:00 UTC is rejected with a validation error. Keep the range entirely before or entirely on or after that date.

2026-02-13

Performance: requests for older date ranges are served from a cluster optimized for historical data, so they respond faster.

2026-02-09

Fixed: the theme filter accepts several themes, as a comma-separated string ("Business,Finance") or an array. Theme names are case-insensitive, and an article is returned if it matches any of them. Before, a theme filter could return no results.
Changed: an unknown theme value returns a validation error. Accepted themes: business, politics, entertainment, sports, finance, crime, tech, general, health, economics, science, automotive, weather, travel, and lifestyle.
Fixed: Search by returned 500 when the request contained only links.
Changed: on Search by, from_ and to_ no longer default to "7 days ago" and "now", so lookups by link, ID, or RSS GUID are no longer limited to the last 7 days.
Search by: embeddings_output is accepted for plans that include embeddings, and responses include geonames.

2025-12-10

Fixed: search_in with translated fields (title_translated, content_translated, summary_translated, title_content_translated) now searches the right fields.
Fixed: when lang held more than one language, only the first search_in field was searched. All requested fields are now searched.
Languages: lang accepts 124 ISO 639-1 codes, up from 55. All previously supported codes still work.