SeekStorm

Twin-core hybrid index Lexical search runs on a sharded inverted index while vector search runs on a sharded leveled IVF index, and a query planner combines them per query with lexical, vector, or hybrid modes and reciprocal-rank fusion.[1][2]

Sub-millisecond lexical latency The project reports about 0.3 ms average lexical query latency over 5M Wikipedia documents in its published benchmark game, with block-max WAND and MaxScore acceleration behind it. Treat this as the vendor's published measurement, not an independent test.[1][2]

BM25F proximity ranking Lexical ranking uses BM25F and BM25F with term proximity, plus six tokenizers including Chinese segmentation, stemming for 38 languages, and optional stopword and frequent-word lists.[1]

Native vector search Vector search supports F32 and I8 precision, cosine, dot-product, and Euclidean measures, TurboQuant and scalar quantization, sentence-aware chunking, and integrated Model2Vec embedding inference alongside externally generated embeddings.[1][2]

Typo tolerance and auto-completion Fuzzy queries with spelling correction return results despite query typos, and typo-tolerant query auto-completion plus instant search back search-as-you-type interfaces.[1]

Facets, geo, and sorting String and numeric-range faceted search with histogram and min/max aggregation, geo proximity filtering and sorting, multi-field result sorting, KWIC snippets with highlighting, and synonyms are all documented.[1]

REST API and six SDKs A CORS-enabled REST API with generated docs is served alongside client libraries for Rust, Python, TypeScript, C#, Java, and Go, plus an adapter that points Algolia InstantSearch frontends at a SeekStorm backend.[1][2][3]

Self-hosted server and cloud The multi-tenancy server runs on Linux, Windows, and macOS with a Docker Hub image, console ingest of CSV, JSON, ndjson, and PDF files, API-key management, an embedded web UI, and an MCP search server; SeekStorm Cloud is the managed alternative.[1][2]

Product teams adding full-text site or documentation search: local file ingest, typo tolerance, facets, and the embedded web UI cover a working search page without building ranking plumbing.[1][2]

Builders doing retrieval over owned documents for AI agents: hybrid lexical plus vector search with an MCP server and REST API fits retrieval-augmented pipelines that stay inside your own infrastructure.[1][2]

Rust and self-hosting shops: an in-process Rust library plus a single-server deployment with memory-mapped indexes suits teams that already operate their own services and want low per-query cost.[1][2]

Teams that will not operate servers: the self-hosted path needs Docker or Rust-crate deployment and index operations, so a moderate effort rating applies and pure no-ops buyers should prefer the managed cloud or another vendor.[1][2]

Buyers needing SSO, compliance paperwork, or a support contract: none of these are documented on the observed official pages, so regulated procurement cannot clear on this evidence alone.[1][2]

Teams choosing by community size: about 1.9k stars were observed on the repository, far below the largest search engines, so third-party guides, plugins, and hiring familiarity will be thinner.[1][2]