Query DSL builder and explainer

Pick an example or start from the form. The Build tab has the form and the JSON side by side: change either and the other follows; anything the form has no editor for stays as JSON and is kept. Below them, the first hits, any aggregations and the important lints update as you type.

The form covers bool (must, should, filter, must_not, minimum_should_match), match, match_phrase, match_phrase_prefix, match_bool_prefix, multi_match (all types), query_string, simple_query_string, term, terms, range, exists, ids, prefix, wildcard, regexp, fuzzy, constant_score, dis_max, boosting, function_score, script_score, nested and knn; paging, sorting, search_after, highlighting, _source and track_total_hits; and aggregations (terms, histogram, date_histogram, range, date_range, cardinality, percentiles, avg/sum/min/max/stats/value_count, top_hits, nested, reverse_nested, global, filter) with sub-aggregations.

Explain describes the request clause by clause, marking query context (scored) and filter context (yes/no, cacheable); shows the Lucene query it becomes (+name:wireless #in_stock:true, (a | b)~0.3, Synonym(…)); lists the lints; has a minimum_should_match calculator (try 2<75% or 2←25% 9←3 and see how many clauses each count requires); and compares from/size, search_after and point in time.

Results lists every hit with its score and highlights — click one for its explanation tree (what _explain returns) and _source — renders the aggregations as bars, and shows the raw response. Sample index shows the mapping and documents, with an _analyze box per field.

Your cluster sends the same request to a real Elasticsearch or OpenSearch with POST _search, _explain/<id> or _analyze — read-only endpoints only — and shows the cluster’s answer beside the toy engine’s.

The toy engine is the same as in the scoring explainer: one shard, BM25 as Elasticsearch computes it, the analyzers of the analyzer playground, minimum_should_match exactly as Elasticsearch parses it (including its float rounding), Lucene’s classic query-string parser (with its a OR b AND c → a +b +c quirk), Lucene’s simple query parser, Elasticsearch date math (now-1d/d, ||+1M, rounding up for gt/lte), sloppy phrases, fuzzy expansion (AUTO = 0/1/2 edits by length, transpositions, blended document frequency), nested documents as hidden documents, and the documented error messages for common mistakes (result window too large, fielddata disabled, unknown query).

Related: Analyzer playground, Scoring explainer, JSON tools, Regex tester, Postgres, Redis.

Links can be written by hand. Add the tab after #: build, explain, results, index or cluster. Requests you build travel in the link’s ?s= parameter; the cluster URL and credentials never do.