Scope: ask. Search uses the shared pipeline rate limit; other operations use the read tier. The key chooses the workspace and sees workspace connections only. No query_id is required. Omitted operation means search. Returns source text, provenance, versions, cost and coverage. Follow next_cursor with next_page_offset, and next_content_offset with source_version for full-source reads. Partial coverage is not evidence of absence. For calendar event browsing, set participant to an email or any.

Lochless workspaces: limit is results per page, default 50, up to 100. The response is results, total_results (every result the query has, across all pages), next_cursor (pass it back as cursor), tokens_scanned and request_id. Only the first page is charged; later pages of the same query are free. Filters (search and browse): from / to (ISO 8601, on the document's own timestamp; from inclusive, to exclusive; a YYYY-MM-DD date alone is that whole UTC day) and metadata ({key: value} exact match; a list value = any of). Only matching documents are read, scanned and billed. dry_run: true (search) returns documents_matched, documents_ready, tokens_to_scan and price_usd without scanning; free, rate-limited like a query. A dry run also returns read_limit_usd and complete (whether that limit reads every matching document).

Read limit (search): read_limit_usd caps what one call may cost, default 1. Over it, the best-matching documents are read first, and only what was read is billed. coverage is {mode, documents_read, documents_total, complete}. When complete is false the response has continue_cursor: send the same request with it to read the next best-matching documents, up to read_limit_usd again; no document is read twice. next_cursor pages stay free. 409 conflict: the documents changed since the first read; run the query again.

Read mode (search): mode is fast or complete. fast reads only the best-matching results (a lookup: one fact or answer), quicker and cheaper, and bills only what it read; it may miss matches. complete reads every matching document, up to read_limit_usd. Without mode the question decides. coverage.mode says which ran; a fast read has no continue_cursor (send mode: complete to read every match).