add_to_listen_later
ChatGPTQueue one to six episodes in the user's Listen Later list in the VO app, as picks from you: each with a one-sentence reason (kept with the pick) and, when a key moment or chapter is the point, a start time (the app shows 'Play from 41:50'). The user's phone gets one notification per batch. Call it when the user's ask already said to add or queue episodes (that ask is the yes), or after the user has seen a proposed list and confirmed it, with the items they kept; never unasked. Episode ids must come from recommend_what_to_hear or search_catalog; VO refuses any other. It only adds: an episode already in Listen Later is left exactly as it is, and nothing in the list is changed or removed. The same batch sent twice within ten minutes is written once. Returns what was added, with vo.fm links.
find_in_listening_history
ChatGPTFind one specific thing the user heard or saved: a claim, a quote, a name, a moment they half remember. Searches only content they actually played or saved, by meaning and by keyword, across their whole history. Each result is dense: the idea in one line, its signal (saved, heard, replayed, possibly_relevant), the episode, when, and a vo.fm link that plays that second. Use when the user wants to retrieve something from their listening. For context while answering a broader question use get_listening_context; for the exact spoken words of an episode use get_episode with include 'words'; for a period use summarize_listening. Optional filters: scope (all, listened, saved), since/until (when the user listened), episode_ids, people. Up to limit items (default 5, max 20), ranked by relevance.
get_episode
ChatGPTOne episode in one call, whether the user heard all of it, part of it or none of it: title, show and duration; an overview of the whole episode, chapter by chapter with a one-line gist and a vo.fm link each, every chapter marked heard or not yet heard, plus where to pick up; how much the user heard; the moments they saved in it (up to 200, each with a vo.fm link); and the listening notes when they exist: the personal recap VO writes after the user finishes an episode, built around their saved moments and its key points. Use it to summarise an episode, to say what the user missed, or to judge whether it is worth a listen. Takes an episode id from summarize_listening (its first entry is the episode playing now), find_in_listening_history or search_catalog. include selects sections (overview, heard, moments, notes, words); the default is the first four. include 'words' returns the exact words the user heard, only the parts they played, with timestamps and vo.fm links, in parts; the result names from_seconds for the next part.
get_listening_context
ChatGPTWhat the user has heard or saved that bears on a topic, for use while answering: three to five dense items, each with the idea in one line, its signal (saved, heard, replayed, possibly_relevant), the episode, when they heard or saved it, and a vo.fm link, plus one coverage line ('heard four discussions of seed financing, saved one moment on running a process'). Use when the user is exploring, recalling or deciding something and their prior listening could give directly relevant context, even if they did not mention VO. Do not use merely because the topic occurs somewhere in their history. Returns no items when nothing strong was found; then say nothing about VO.
recommend_what_to_hear
ChatGPTWhat the user should hear next, in one call. Give the user's ask in plain words; VO builds the searches from it and from their listening profile, leaves out episodes they finished (or heard 80% of), keeps partly heard ones flagged with how much they heard and which key moments they have not heard yet, marks how new each option is to them ('new to you', 'near what you heard', 'on a topic you saved'), and returns up to limit options (default 20) best match first, each marked strong or weak and with what it matched (the ask, the profile, or both), with show, length, date, a snippet, up to two key moments with start times, and whether it is already in Listen Later. The reply also carries what VO knows about the user (recent listening), so no other call is needed first. VO does not rank the options for you: you choose the picks. When all_weak is true, VO found nothing that fits: say so. Add queries only for angles VO cannot know from this conversation. Not for finding one known episode: that is search_catalog. It only reads: it adds nothing to Listen Later and changes nothing the user can see; adding is add_to_listen_later, after the user says yes.
remove_from_listen_later
ChatGPTTake one episode out of the user's Listen Later list in the VO app, typically a pick they did not want ('drop the second one'). Only the list entry goes; listening, saved moments and notes are untouched. The episode id must be exactly as a tool returned it. Says whether it was in the list and whether it was a pick.
search_catalog
ChatGPTFind one known episode or show in the VO catalog by name, host or guest, when the user names it ('the Acquired episode on Nvidia'), or the fallback when the user names the episode they are playing and summarize_listening does not show it first. Returns episodes and podcasts with ids for get_episode and add_to_listen_later, and vo.fm links. Not for topic browsing or recommendations: that is recommend_what_to_hear. Not for what the user heard: that is find_in_listening_history. types can add key moments, the clips VO cuts inside catalog episodes (the same for everyone, not the user's saved moments). Up to limit results (default 5, max 20).
summarize_listening
ChatGPTWhat the user listened to and saved, and a recap of it: a timeline of every episode heard in a window (title, show, how much heard, when, vo.fm link), the listening notes and saved moments of the most significant ones, and, with include 'listen_later', what is in their Listen Later list with the picks an assistant added. Use for 'what did I listen to', 'recap my week', 'what is in my Listen Later'. Call it first, with no window, when the user names an episode they are playing ('I've just started', 'I'm 12:30 into', 'I'm finishing'): the timeline opens with the episode playing now or played most recently, with its episode_id for get_episode. Times are when they listened. With no window, the whole history as a timeline only. For one topic use find_in_listening_history; for one episode use get_episode. The details may be a subset of the window that fits max_chars, and the result states how many of how many; topic steers which episodes get detail. Lists are paged with limit (default 50, max 200) and offset; the result names next_offset when more remain.
get_episode
Claudeget_heard_transcript
Claudeget_memory_digest
Claudelist_listening_history
Claudelist_saved_moments
Claudesearch_catalog
Claudesearch_memory
Claudewhoami
Claude
