The engine

Lucenia is a distributed search and analytics engine on an Apache Lucene core. Four retrieval modalities share one index, one query language, and one relevance computation — so hybrid queries are native, not stitched.

query surface

Field types and queries

As documented at docs.lucenia.io. No roadmap items in this table.

ModalityField typesQueries
Full-text text keyword matchmatch_phrasematch_phrase_prefixmatch_bool_prefixmulti_matchquery_stringsimple_query_stringintervalsfuzzyspan
Vector knn_vector approximate k-NN (faiss, nmslib) · exact k-NN via script score · filtered k-NN · nested-field k-NN · vector quantization · painless distance functions
Geospatial geo_point geo_shape geo_bounding_boxgeo_distancegeo_polygongeo_shape
Cartesian xy_point xy_shape xy_shape with the full predicate set, operating in pixel space
Structured numeric · date · range · object · nested termtermsterms_setrangeexistsprefixwildcardregexpids
Aggregations bucket · metric · pipeline
Compound booldis_maxfunction_scoreboostingconstant_scorescript_score
operations

Built for production clusters

AreaDetail
Ingestingest pipelines with processors · bulk API · data streams · Logstash
Index managementtemplates · aliases · rollups · transforms · index state management · codecs
Search at scaleconcurrent segment search · asynchronous search · point-in-time pagination · cross-cluster search
Relevance toolingsearch pipelines · rank evaluation · query profiling · side-by-side comparison in Dashboards
Resiliencesnapshot/restore to object storage · cross-cluster replication · shard allocation awareness
Observabilityquery insights · performance analyzer · audit logs · Dashboards
migration

If you run OpenSearch or Elasticsearch today

Lucenia’s API surface, Query DSL, and client model follow the conventions your applications already use. Existing queries, mappings, ingest pipelines, and dashboards carry over with minimal change, and official clients are available for Java, Python, Go, and JavaScript. Migration guides are in the documentation.

benchmarks

Measured, not asserted

Results are produced with OpenSearch Benchmark against a reproducible harness, published alongside the numbers.

Workloadp50p90p99vs. baseline
geo_shape INTERSECTS · 100M docs
hybrid BM25 + k-NN, filtered
k-NN recall@10
mounted Parquet, cold query
cost per queryable TB / month

Publication pending — the harness ships with the results.

Bring us a query your stack can’t answer