Node reference

Query

Read stored data back with a filter or a plain-language question.

Read-only. Three ways to ask:

SELECT * FROM llm_results WHERE name LIKE '%Acme%' LIMIT 50

Collections: raw_pages, pages, llm_results, records.

Mongo filter: collection + mongo_filter.
Plain language: a prompt such as “which companies mention solar?”.

Scope

Setup	Reads
Query alone	Your whole database, across stored runs
Downstream of a Data Source	Only that run’s saved documents
`source_execution_id`	A previous run (narrow further with `source_node_ids`)

Feed a fetcher from stored data

Connect a Query to a URL Fetcher with url_source: query_result and url_field (e.g. url). The fetcher crawls the URLs your records contain. Use this to re-visit or enrich earlier results.

Input, output, and limits

No upstream node is required for a database-wide query. A connected Data Source narrows the scope to saved documents from that run. Results are emitted on query_result; connect Table to inspect them or URL Fetcher to revisit their URLs.

{"collection":"records","mongo_filter":{},"limit":20}

limit is from 1 to 100; sort accepts up to two field/direction pairs and fields up to 50 field names. Prefer deterministic queries when the filter is already known. Plain-language queries can require a model. Validation or execution rejects unsupported query syntax, inaccessible source runs, and unavailable stores. See configuration types.