Query DSL

Nixiesearch has a Lucene-inspired query DSL with multiple search operators.

Note

To search over a field, make sure that this field is marked as searchable in index mapping.

Unlike Elastic/OpenSearch query DSL, Nixiesearch has a distinction between search operators and filters:

• Search operators affect document relevance scores (like semantic and match)
• Filters only control how we include/exclude documents. See Filters for more details.

Search request format

Search request format is similar to existing Lucene-based search engines:

{
  "query": {
    "match_all": {}
  },
  "fields": ["title", "desc"],
  "size": 10,
  "aggs": {
    "color_counts": {"term": {"field": "color"}}
  },
  "filters": {
    "include": {"term": {"category": "pants"}}
  }
}

Where fields are:

• query: required, a search query operator.
• fields: optional (default: all stored fields), which document fields to return in the response payload. Note that these fields should be marked as store: true in index mapping.
• size: optional (default: 10), number of documents to return
• aggs: optional, facet aggregations, see Facets for more examples.
• filters: optional, include/exclude filters to select a sub-set of documents for searching.

Search operators

Search operators allow you to actually perform the full-text search over your documents. They're designed to be fast and quickly get top-N most relevant document for your search query.

Nixiesearch supports the following search operators:

• match: search over a single field
• multi_match: search over multiple fields at once.
• match_all: match all documents.
• semantic: embed a query and perform a-kNN vector search over document embeddings.
• knn: perform a-kNN vector search over document embeddings (without embedding the query).

Operators can be combined into a single query:

• dis_max: search over multiple fields, but sort by the most matching field score.
• bool: combine multiple queries in a boolean expression.

Note

All search operators can be combined with filters to search over a subset of documents.

Ranking operators

Rank operators accept one or more search operators but only operate on top-N of them.

Nixiesearch supports the following list of rank operators:

• RRF: Reciprocal Rank Fusion, merge two search results lists based on document position.
• Cross-Encoder: Neural reranking model that jointly processes query-document pairs for more accurate relevance scoring.