smry

Remove one item from this account's library. Web sources are unaffected. For an uploaded file, its original, extracted text, and synced local copies are permanently deleted; highlights and notes are kept.

remove_library_item

Save supplied source text without fetching a public URL. Use external_id for retry-safe creates; content_kind can correct automatic classification.

save_content

Securely extract a public URL and idempotently add it to the account's library. Returns alreadySaved=true for a duplicate and reports hard paywalls instead of claiming success.

save_article

Search and paginate across the account's saved items, standalone notes, and highlights through one corpus. Uses the same Postgres matching as the app and REST API; approximate=true identifies spelling recovery. Returns stable refs for get_library_object. Use list_library when you specifically need item lifecycle filters or reading history.

list_library_objects

Return only the stored passages relevant to a question about one library item. Use after list_library instead of reading a long book, PDF, email, transcript, or document from the beginning.

search_saved_item

Search this account's articles, websites, books, emails, PDFs, social posts, videos, podcast episodes, and documents. For a normal text search, pass only query and limit. "Triage my inbox" is {"status":"inbox","limit":10}; "saved this week" is {"saved_after":"7d"}. Matching is shared with app search and REST, including spelling recovery (approximate=true). Optional filters are combined with AND: omit every filter the user did not explicitly request. Filter by exact tag, kind, capture method, format, or lifecycle; set view=history for recently read items. Items are ordered by most recent status change first (history view: last read first); continue with nextCursor. Item IDs select later tools and are not smry.ai routes; cite only each exact returned source URL.

list_library

Discover current public sources with up to four focused searches. Search snippets are discovery aids; open the strongest pages before citing substantive claims.

search_web

Preview changes to matching sources, entries, or collections. Default dry_run:true returns would_affect, sample, and commit_token without changing them. Commit by sending the same op and commit_token; retries replay the recorded result. Explicit dry_run:false applies immediately. sources.move needs set.collection; sources.rename needs set.title; collections.create/rename need set.name. Collections rename/delete need exactly one where.collection. Use manage_collections for direct collection CRUD. Check partial and each outcome after committing. Example: {"op":"collections.create","set":{"name":"Research"}} Example: {"op":"sources.move","where":{"source":{"include":[42]}},"set":{"collection":"Research"}} Example: {"op":"sources.move","commit_token":"<commit_token from preview>"}

mutate_feeds

Less like this hides an included entry. Feedback remains separate agent context and never silently revises the Collection filter. Example: {"entryId":42,"value":"less_like_this"}

set_feed_feedback

Read the append-only library change stream, including deletions. For a race-free first sync, call once with cursor=latest, list the library, then continue from the returned nextCursor.

get_library_changes

One stream, two checkpoint owners. Stateful callers pass cursor:"latest", then store each next_cursor themselves; that first call only establishes the checkpoint and returns no events, so poll again with the returned next_cursor. Stateless scheduled agents pass consumer_id; smry retains the acknowledged position and the consumer's saved where predicate. Pass exactly one. Managed batches repeat until their next_cursor is acknowledged. Every response also carries a capped source-health attention block. To read what is already there rather than what changes next, use query_feeds. Example: {"cursor":"latest","limit":20} Example: {"consumer_id":7,"limit":20}

get_feed_changes

list: Returns the same owner-scoped Collections shown in /feeds, including canonical filter state and revision. Required fields: none. Example: {"action":"list"} create: Creates the same named Collection used by the production /feeds UI. Required fields: name. Example: {"action":"create","name":"Research"} update: Membership is a complete replacement, matching the production Collection editor. Required fields: id. Example: {"action":"update","id":12,"name":"Engineering"} delete: Deletes a custom Collection and moves its inputs to Unsorted. Required fields: id. Example: {"action":"delete","id":12}

manage_collections

create: Save a where predicate and let smry retain one processing checkpoint. New consumers start at the current high-water mark, so they receive future discoveries rather than replaying history. For history, create the Consumer first, query the historical window separately, then poll; that order closes the query/create race. Create one per independent processing loop, not per source. Retrying the same case-insensitive name and predicate returns the existing Consumer; the same name with another predicate is a 409. Required fields: name. Example: {"action":"create","name":"Engineering briefing","where":{"collection":["Research"]}} list: List each saved consumer, its opaque acknowledged_cursor, delivery and acknowledgement timestamps, and how many matching entries are waiting past that checkpoint. Required fields: none. Example: {"action":"list"} ack: After processing every event returned by getFeedChanges for this consumer, acknowledge that response's next_cursor. Crashes before ack cause safe redelivery; stale, foreign, and invented cursors are rejected. Required fields: id, cursor. Example: {"action":"ack","id":7,"cursor":"<next_cursor from get_feed_changes>"} delete: Remove only the saved predicate and checkpoint. Followed sources, entries, and human read state are untouched. Required fields: id. Example: {"action":"delete","id":7}

manage_feed_consumers

preview: Classifies a bounded recent sample and returns both binary decisions, uncertainty counts, reasons, and degradation status. Save with applyFeedFilter using this preview token and Collection revision. Required fields: collection, filter. Example: {"action":"preview","collection":{"id":12},"filter":{"prompt":"Software engineering articles","whenUncertain":"include"}} apply: Atomically applies the exact preview when the Collection revision still matches. Passing a preview created with filter:null clears the filter. Required fields: previewToken, expectedRevision. Example: {"action":"apply","previewToken":"<previewToken from preview>","expectedRevision":0}

manage_feed_filter