Stasht

Read saved text, transcripts, metadata and selected images to answer questions about an item. Use returned stable refs for stash, entity, collection or tag, including their context witness when supplied. Unknown, removed and inaccessible IDs have the same status. Thin saves can still be available. Personal titles, notes and save dates belong to this account; source and event dates describe their respective objects. Stored transcripts provide what was said; image pixels support visual inspection. Check image_status and delivered_image for actual delivery. A video cover shows only that cover, not unseen frames. Media indices are positions in this response, not permanent identities. Evidence sections distinguish source text, OCR, transcripts and accumulated entity metadata. Their inventory and continuations allow further reading; absent extraction does not establish that the source lacks the information. Use show_stasht_items to show image cards. The person can open a card in Stasht to view its media; this tool supplies evidence for your answer.

Discover collections, tags and a recent overview of the person's saves. Catalogs supply stable refs and current role/state. Same names can have different IDs; preserve ambiguity. Tags use account last-used recency; use name_query to find names; duplicate names retain their separate IDs. Suggested tags are excluded unless include_suggested=true, and labeled. Collections default to owned; request collection_access=shared or all explicitly to include active shared memberships. Invitations are excluded. Read collection/tag details with get_stasht_items; enumerate members with search_stasht(scope=ref). Overview counts describe only its stated recent window. Explore relevant saves to support a broader interpretation of interests.

Identify the Stasht account linked to this connection.

Find remembered sources or enumerate a precise library population. Choose searches that fit what the person means. Semantic search discovers relevant saves; text search finds indexed words and phrases. Names can be useful in either. Related places, concepts and terms can help when initial results cover only part of the request. For more suggestions, explore a useful aspect of the request and avoid repeating items already shown. For more results from the same text search or scope, follow next_cursor with identical arguments. Text search is a different population, not the next page of semantic results. query_mode=semantic returns a ranked sample with no total or continuation. query_mode=text with order=default supplies indexed_match_count and paging in stable identity order. That count describes indexed text matches, not every conceptually relevant save. limit is page size; returned_count counts this page. auto uses semantic recall for an unscoped query and enumeration for scoped/chronological reads. Discover personal tags separately with get_stasht_library(name_query=...), then enumerate the tag scope. Optional save-time bounds must include UTC or an offset; after is inclusive, before is exclusive. These are the person's latest save times, not event or publication dates. Empty query, kind=stash lists recent independently saved sources; kind=entity lists explicit Stuff saves, including entity-only accounts. Resolve collection/tag IDs with get_stasht_library, then pass that ref as scope. A collection supports kind=all (mixed members); tags select stashes. A stash scope with kind=entity lists its mentioned concepts. An entity scope lists authorized connected sources/entities; these are not all explicit saves. Preserve returned context witnesses when passing refs to subsequent tools. Follow next_cursor with identical arguments; empty pages can still have more. Default scope order follows its index. Newest/oldest uses the stated clock, ordering the complete scope before limiting (five pages, at most 500 examined rows including removed states); unknown clocks follow dated items. A text query with newest/oldest filters the save-clock enumeration and may need further pages to fill a response. Ranked semantic recall has no chronological paging. Scoped query is lexical matching in inspected fields, not semantic ranking. near requires explicit coordinates and radius_km, never assumed GPS. Geographic and event constraints apply to the same entity (kind=entity). event_from is inclusive, event_until exclusive, interpreted in time_zone. Missing location/date metadata excludes only that constrained projection. Start from the person's intent; use returned evidence to develop useful connections and tentative suggestions. Ranked samples do not establish totals. Use get_stasht_items to inspect selected evidence and media; Stasht URLs are for the person to reopen the save, not a required web-fetch step.

Show selected Stasht items as image cards with links that open in Stasht. Pass stable refs selected by search/detail. Each card opens the saved item in Stasht in a new tab. Read evidence and inspect media with get_stasht_items for the conversation. The view never saves or edits. Hosts without MCP Apps can use the returned Stasht links.