Ranked search over one public table's rows, the same search its dataset page runs: words, prefixes, typos and spelling correction, served from the table's search package. NOT a catalogue search (search_datasets finds datasets) and not SQL. Example: {"table_id": "0f2f...", "query": "blade runner 1982"}. settled defaults to true (a finished query, which may be planned by a model); send settled: false for a fast keystroke answer planned by the parser alone. page_size is 1-100 (default 20). columns is a subset of the table's search display columns (default all of them). Returns {rows: [{row_id, fields}], plan, planner, retrieval, next_cursor, count, package: {generation_id, package_digest}, table_version_id, timings}. For the next page send the same query with the answer's plan and next_cursor. like: a row_id from an answer finds the rows most like that row (more like this; query, when not empty, is a topic beside it), on a table whose search has meaning vectors. mode (first page only): auto (default) lets the planner choose; meaning ranks by meaning alone; image finds rows whose pictures look like the query (posters showing a red sky over a city); emoji searches by the query's emojis. Most tables search words only. like and the meaning and image modes need meaning vectors, which a table has only when capabilities.search.semantic_columns from get_table_schema is not empty. When it is empty, use the default mode. A table whose search cannot run a mode answers invalid_plan. count: true also asks for the exact number of matching rows; count is null when the table cannot count yet. corrected, when present, is the spelling the search used. A table whose search collapses (a passages table that answers for the entities of another table) returns one result per entity: the answer carries collapsed_by (the key column), and each row is the entity's best match with matched_via (up to 8 of its other matching rows, each {row_id, fields}, in ranked order) and group_size (how many of its rows matched in the ranked window). Paging moves over entities. group: "row" asks for every matching row on its own instead; send the same group with a later page. A table whose search does not collapse accepts group and answers ungrouped rows. To point search at a passages table, in this order: declare collapse in its search spec; wait until its search has been rebuilt and is active for that declaration (a newly declared collapse reaches search only after its search build finishes and activates); run one search on that table and confirm the answer carries collapsed_by; only then set search_table_id. If the answer does not carry collapsed_by, do not set search_table_id. Find the table with search_datasets and list_tables. A table has row search only when get_table_schema returns capabilities.search. Its state says whether search_table answers now. ready means it answers. preparing means a package is being built, and search may still answer from the previous version's package, so try it. unavailable means no package and no build. No state means Cloud has not read it yet, or the table is a recorded stream, whose state Cloud never reads, so try it. A refused search answers search_unavailable with a reason and retry_after_seconds. preparing means the search package is being built or loaded, and service_unavailable means the search service is down. For both, wait retry_after_seconds and retry once. build_failed means the package failed to build, no_package means the table has no search package, and not_public means another workspace's private or link-shared table. For these three, retry_after_seconds is null and retrying will not help, so use query_table or sample_rows instead. Your own workspace's tables are searchable whatever their visibility. To filter, sort or aggregate, use query_table. No key needed on a public table: a keyless caller gets 30 calls a minute and 600 a day per address, and 300 a minute and 10,000 a day across all keyless callers, and settled calls share the dataset page's per-table limit. A refusal past that answers rate_limited with retry_after_seconds. Without a key, private and link-shared tables, and public tables whose publisher turned download off, answer table_not_available. An mr_use_ workspace key (Authorization: Bearer) or an OAuth connection lifts the limit to the key's query limit (60 calls a minute per key or connection, shared with query_table, and 5,000 a day per workspace) and adds the workspace's own tables. A keyed call counts against the workspace's allowance, as query_table does. Cite the package generation and digest.