Search Threads without a login: keyword and tag searches, top or recent posts, or one page per date window down to one hour for depth. One row per post with text, author, time, likes, replies, reposts, media, language and whether the text contains your words. JSON, CSV, API. Pay per post.
0.1 maintenance build (2026-10-11): fixes shared with the other Threads listings, no price, input or output change
A GraphQL answer that carries thread items without one post next to execution errors (a persisted query id the site no
longer pairs with these provider flags) ends the subject as unexpected_answer with the reason in the coverage row and
leaves it open for a restart; it was read as the end of a feed, a silent short delivery (PENDING 231 (a)).
The monitor memory keeps a backlog record for profile walks: a walk cut by a cap, the charge limit, the work allowance
or a failed call records the feed cursor it did not deliver in full, and the next run under the same memory, after
reading the known newest posts (recognised by the post codes the record keeps, since the site gives the same position
a different cursor on every run), jumps to it and continues; a run that reads to the end clears it (PENDING 231 (c)).
The memory summary shows backlogRecords. The page guard of 2,000 feed pages is a cut (page_guard), not the end.
Status texts no longer say "nothing was charged" for a whole run that bills the run start event (PENDING 231 (b)).
0.1 (2026-10-07)
First build (0.1.8, 2026-10-07): profile posts (profile row plus the feed with its cursor), keyword and tag search (top, recent, daily windows), replies under a post, filters in code (dates, languages with unknown kept, likes, replies, media), cross run memory, one sticky datacenter proxy session per run.