LeanCorpus

A fast, embeddable full-text search engine for modern .NET. Zero dependencies, AOT-ready, segment-centric design.

dotnet add package LeanCorpus

Five-minute quick start

using Rowles.LeanCorpus.Document;
using Rowles.LeanCorpus.Document.Fields;
using Rowles.LeanCorpus.Index.Indexer;
using Rowles.LeanCorpus.Search.Queries;
using Rowles.LeanCorpus.Search.Searcher;
using Rowles.LeanCorpus.Store;

// Create an index
using var dir = new MMapDirectory("./my-index");
using var writer = new IndexWriter(dir, new IndexWriterConfig());

// Add documents
var doc = new LeanDocument();
doc.Add(new TextField("title", "The quick brown fox"));
doc.Add(new StringField("id", "1"));
writer.AddDocument(doc);
writer.Commit();

// Search
using var searcher = new IndexSearcher(dir);
var hits = searcher.Search(new TermQuery("title", "fox"), topN: 10);

foreach (var hit in hits.ScoreDocs)
    Console.WriteLine($"docId={hit.DocId}  score={hit.Score:F3}");

Why LeanCorpus


Zero allocations in the hot path	Ref-struct token pipeline. No per-token heap allocation during analysis or scoring.
Modern scoring	BM25 (default), BM25+, BM25L, TF-IDF, Dirichlet, Jelinek-Mercer, and SIMD-accelerated cosine. Block-Max WAND for sublinear top-k.
Vector search built in	HNSW graphs with pre/post-filtering and BBQ quantisation (32x compression). Not a plugin.
Native AOT	Trim-safe, publish-ready with `PublishAot`. Validated by a dedicated smoke-test suite.
LINQ queries	`IQueryable<T>` provider translates expression trees to LeanCorpus queries.
Observability built in	OpenTelemetry tracing and metrics, slow-query log, Aspire dashboard integration.
Competitive performance	Benchmarks against Lucene.Net 4.8 on 100k-document corpora. Up to 63x faster highlighting, 33x faster geo-distance queries, with 10x to 100x less allocation.

Start here

	Start with this
Build your first search	Create an index, add documents, and run a query in three short steps.
Choose a package	Find the core library, standalone Rowles.Text, source generator and compression packages below.
Learn the architecture	Understand documents, analysis, segments, readers, queries and vectors in Concepts.

Is LeanCorpus right for me?

Choose LeanCorpus when you need embedded, local .NET search with control over index files, query execution and Native AOT deployment. Choose Lucene.NET when its mature ecosystem and broad compatibility matter more than a modern .NET-first API. Choose Elasticsearch when you need a distributed service, operational tooling and multi-node scale. A database full-text feature is often simpler when search is secondary to transactional data and the required query behaviour is modest.

Feature tour

Analysis

Zero-allocation span-based pipeline: tokenisers, 28 token filters, 11 language stemmers, synonym graphs, Hunspell dictionary stemming, ICU, CJK tokenisation, pattern tokenisers, phonetic filters (Metaphone, Beider-Morse style). Browse analysis docs

Search

25 query types including term, boolean, phrase, fuzzy (Myers bit-parallel), regexp (FST automaton), span, intervals, block-join, geo, and vector. Browse search docs

Scoring

BM25, BM25+, BM25L, TF-IDF (classic plus three variants), three language-model similarities, and SIMD cosine. Pluggable similarity API with expression-free function scoring. Browse scoring docs

Concurrency

Multi-threaded indexing with per-thread DWPT pools. Near-real-time readers via SearcherManager. Background merges that never block commits. Lease-protected segment caches. Browse concurrency docs

Observability

DefaultMetricsCollector, MeterMetricsCollector (System.Diagnostics.Metrics), OpenTelemetry tracing, structured slow-query log, Aspire dashboard integration. Browse observability docs

Extensibility

CodecKit for custom storage formats. Pluggable compression (LZ4, Snappy, Zstandard). Source-generated document schemas. Index codec migrator for format upgrades. Browse contributor docs

Packages

Package	NuGet	Description
LeanCorpus		Core library. Zero dependencies.
Rowles.Text		Standalone tokenisers, filters, stemmers and analysers.
LeanCorpus.SourceGen		Roslyn source generator for typed document mapping
LeanCorpus.Compression.LZ4		LZ4 stored-field compression
LeanCorpus.Compression.Snappy		Snappy stored-field compression
LeanCorpus.Compression.Zstandard		Zstandard stored-field compression