GET /v1/search asks all twelve networks for a keyword and returns one merged, ranked list with a per-network total. Underneath, the queries that took 40 to 60 seconds on Reddit and Bluesky now take under a second: text indexes for keywords, windows on the post date, and pages picked before rows are read. time_range takes any number with a unit and refuses what it does not understand. Pinterest rows carry the pin id, its URL and its image.
An agency customer sent us his test log: Pinterest rows with no pin reference, most networks empty for his keyword, and some calls taking 43 seconds. All three were ours. This release is the answer.
sources=pinterest,x,reddit), one answer: each network returns its top rows, normalized to one shape (platform, title, post_url, thumbnail_url, author, engagement, stats, date, and the network's full row under raw), merged and ranked by engagement or date. sources[] says how many posts matched on each network, so you can see where a topic lives and where we hold nothing. Docs at /api/v1/docs.php#search.pin_id, post_url and image_url.meta.total reports 10001 beyond that; has_more stays exact. Counting every match of a broad window cost as much as the page.The radar in the app runs on the same query paths and got the same fixes. The skill and the docs describe the new endpoint; nothing changed in existing parameters.