Parse every link out of pasted text (a KakaoTalk or Slack export, a bare list, anything), skip what is already archived, extract YouTube transcripts, and prepare a page payload for each one. Returns a batch ID immediately; extraction runs in the background. Poll get_batch_status.

Progress for a running or finished batch. When state is 'completed', ready_to_save holds one payload per item that is ready to write to the workspace, and items explains anything that failed or was skipped.

Pull every URL out of a text blob and classify it by platform, without archiving anything. Useful for showing the user what was found first.

Get the transcript for one YouTube video, trying caption APIs, yt-dlp, and local speech-to-text in turn. Reports which method worked and why the others did not. Cached, so asking twice is free.

Write summaries and tags for items a batch has extracted but not yet summarised, using your own model. Call this after archive_batch reports items in needs_summary, and call it repeatedly until remaining reaches zero. Works a few items at a time on purpose: a thousand transcripts cannot go through one request.

Record a summary and tags for one extracted item. Use this after summarise_pending hands you a prompt to write yourself. The summary is checked for copying the source before it is kept, and the item then appears in ready_to_save.

Full-text search across everything archived — transcripts and article bodies, not just titles. Use this to answer questions about what the user has already saved, or to find the source of a half-remembered idea. Returns matching excerpts with the workspace page link where one was recorded.

Has this URL already been archived? Matches the canonical form of the URL, and if a title is supplied also compares it against archived titles to catch re-uploads and mirrors.

Tell Fetchive that a URL has been saved to the workspace, and where. Call this after every successful save — it is the only way duplicate detection learns what exists.

Remove a URL from the archive index, so it counts as new again. Use when the page it points to was deleted from the workspace. Does not touch the workspace itself.

Shape one link into a page payload: properties, the workspace's existing tag vocabulary, and instructions for classifying against it. Quotes are capped and transcripts are never included verbatim.