@seekite/build
v0.1.0-alpha.0
Published
Shared indexing pipeline for Seekite
Maintainers
Readme
@seekite/build
The shared Seekite build engine: sources, extraction, heading-aware chunking, lexical indexing, embeddings, and portable static serialization.
import { buildSearch, defineSearch, defineSource, staticVectors } from "@seekite/build"
await buildSearch(defineSearch({
corpora: {
docs: {
source: defineSource("api", async () => fetchDocuments()),
vectors: staticVectors(provider),
},
},
}))Static vectors are cached under .seekite/cache by provider ID and exact
embedding text. Pass { cache: false } to bypass reads while still warming the
cache, or { cacheDir: "/mounted/cache" } to override its location. The
SEEKITE_CACHE_DIR and SEEKITE_NO_CACHE=1 environment variables provide the
same controls for CI.
The builder writes sharded format v2 with int8 vectors by default. Pass
staticVectors(provider, { quantize: "f32" }) for f32 output, or
buildSearch(config, { format: 1 }) for the temporary legacy escape hatch.
Framework adapters share loadSearchConfig() and runIntegrationBuild().
The latter loads search.config.ts when present and otherwise indexes a default
docs corpus from rendered HTML. generatedHTML() accepts ordered
bodySelector values plus excludeSelectors; defaults prefer
[data-seekite-body], main, and article, while excluding navigation,
sidebars, scripts, SVG, MathML, and [data-seekite-ignore] content.
