Extract Google Trends data: daily trending searches, keyword interest over time, related queries and topics, regional interest. 150+ countries, category filters, Google property selection (Web, YouTube, News, Images, Shopping). From $1 per 1,000 trending results.
Trending mode no longer uses BasicCrawler + RequestQueue. Per-request queue storage cost (~$0.000018/write × 4 writes per Request) dominated the sub-second RSS fetches and made daysBack=90 ~80% more expensive than v0.2 despite being faster. Replaced with a lightweight promise pool (mapWithConcurrency). Result: T_HUGE (daysBack=90) now ~25% cheaper than v0.2 at 2x the speed.
exploreConcurrency input (1-5, default 3). Parallel Playwright pages on a shared browser context.
Changed
Trending mode now fetches days in parallel via a lightweight promise pool (no RequestQueue). Sub-second RSS fetches don't justify BasicCrawler/RequestQueue per-request storage overhead.
Explore mode now runs N keywords in parallel across a pool of Playwright pages (single shared browser, single cookie warmup) via BasicCrawler. Within each keyword, the 4 widget API calls (multiline / comparedgeo / relatedsearches × 2) now run in parallel via Promise.all.
Charging adopts the charge-first/push-after pattern so budget-capped runs no longer leak one paid-for row past the cap.
Removed
TRENDING_DAY_DELAY_MS (300 ms serial delay between days). Concurrency replaces it.
EXPLORE_DELAY_MS (500 ms serial delay between keywords). Concurrency replaces it.
[0.2.0] - 2026-04-27
Added
language input (16 locales: en-US, en-GB, pt-BR, pt-PT, es-ES, es-MX, fr-FR, de-DE, it-IT, ja-JP, ko-KR, zh-CN, zh-TW, ru-RU, hi-IN, ar-SA). Applies to both trending and explore modes.
daysBack input (1–90) for trending mode — fetches RSS for the end date plus N-1 prior days.
date input (YYYY-MM-DD) for trending mode — end of the daysBack window. Defaults to today in the chosen timezone.
timezone input (IANA names, ~22 supported) for trending mode.
Trending output now includes date and language fields on every record.
README "MCP / Agent Usage" section with a verified multi-parameter invocation.
Changed
Default tz for explore mode is now 0 (UTC) instead of the prior hardcoded 240 (US-Eastern DST). Records near midnight may now be assigned to a different calendar day. Pass timezone=America/New_York for prior behavior.
maxResults semantics: with daysBack > 1, the cap applies per day, not total.
maxResults schema range narrowed to 1–50 (default 10), down from 1–200 (default 50). Google's trending RSS returns up to ~10 items per day per country — higher values had no effect and the documentation now reflects reality.
Fixed
Historical trending with daysBack=1 and an explicit past date no longer silently returns today's data. The RSS URL now always includes the date query param so the snapshot endpoint is hit consistently.
The actor no longer FAILs on user-input mistakes. Per project policy, Actor.fail() is reserved for unexpected internal errors. Invalid inputs (bad language, timezone, geo, date, daysBack out of range, etc.) are auto-fixed to safe defaults with a warning logged. Zero-result runs now SUCCEED with a warning explaining why and the user is not charged. This protects the store-displayed success rate.
Performance
Lowered defaultMemoryMbytes from 1024 to 512, minMemoryMbytes from 256 to 512, maxMemoryMbytes from 4096 to 2048. Real peak memory observed in production: 120–237 MB across both modes. The 1024 MB default was charging users ~4× more in compute units than needed; the new 512 MB default still gives 2× headroom over peak. Net effect: default-tier users pay ~half the compute cost for the same data.
Internal
Added node:test test runner via pnpm test.
Added pnpm typecheck script.
Refactored URL-building to pure functions for testability.
[0.1.0] - 2026-04-25
Added
Initial release split out of the bundled apify/actors/ repository into its own repo under the scrapeforge organization.
pnpm as the package manager (pinned via packageManager field).
