I'm building LeanLucene, a low-allocation alternative to Lucene.NET for indexing and search

Over the last few months I've been building LeanLucene, a modern alternative to Lucene.NET. It targets .NET 10 & 11, and has one external de

#clippings

NuGet Gallery | LeanLucene 1.1.1

Over the last few months I've been building LeanLucene, a modern alternative to Lucene.NET. It targets .NET 10 & 11, and has one external dependency (NativeCompressions for LZ4 and Zstandard), everything else is BCL and modern C#. Benchmarks are looking good at the moment, with nearly all being > 50% better, and > 99% better in some.

So far I've implemented

  • Segment-centric indexing as immutable segments flushed to disk, background merges, atomic segments_N commits
  • Memory-mapped reads with a MMapDirectory
  • BM25 scoring by default (TF-IDF available); BlockMaxWAND for early termination across segments
  • BKD trees for numeric range queries; HNSW graphs for KNN vector search
  • Full analysis pipeline. Stemmers for 12 languages, N-gram/edge N-gram/CJK bigram tokenisers, synonym graph, accent folding
  • Query parser supports field:term, phrases, slop, fuzzy, prefix, wildcard, +/-, grouping, and boosts. Ex. +title:lean "full text"~2 fuzzy~1 prefix* field^2.0
  • Facets, histograms, field collapsing, highlights, DidYouMean suggestions
  • Block-join indexing for parent/child documents
  • Geo queries (bounding box, distance)
  • Schema validation, deletion policies, point-in-time snapshots
  • Optional OpenTelemetry traces + metrics, slow query log, search analytics
  • ASP.NET Minimal API example (multi-collection HTTP search service)
Built with LogoFlowershow