Indexing primitives: pluggable full-text and vector indexers (in-memory, DuckDB, PGlite).
| Package | Description | Published |
|---|---|---|
| @statewalker/indexer-api | Pluggable indexer contract: full-text, vector, hybrid. | yes |
| @statewalker/indexer-chunker | Markdown-aware chunking utilities. | yes |
| @statewalker/indexer-core | Workspace-internal scaffolding consumed by the backends (composite index, merge, SQL retrievers, generic factory builders). | no |
| @statewalker/indexer-mem | In-memory vector sub-index (MemVectorIndex) used by the FlexSearch/MiniSearch indexers. |
yes |
| @statewalker/indexer-mem-flexsearch | FlexSearch + MemVectorIndex + optional persistence. |
yes |
| @statewalker/indexer-mem-minisearch | MiniSearch + MemVectorIndex + optional persistence. |
yes |
| @statewalker/indexer-duckdb | DuckDB backend: real BM25 FTS (fts extension) + HNSW cosine vector (vss extension). |
yes |
| @statewalker/indexer-pglite | PGlite backend: tsvector/GIN FTS + pgvector HNSW cosine. |
yes |
| @statewalker/indexer-search | Workspace-internal app-side search-orchestration stack (SearchPipeline, embed helpers, reranker blending, mocks). QMD-port utilities live under ./utils. |
no |
| @statewalker/indexer-tests | Shared Vitest conformance suite run by every backend. | no |
This repository depends on:
| Repository | Packages used |
|---|---|
statewalker-db |
@statewalker/db-api |
Cross-repo dependencies are declared workspace:* rather than catalog:. This is
deliberate: turbo derives its task graph from workspace: specifiers and does not
resolve catalog:, so a catalog: cross-repo dependency is invisible to the scheduler
and its consumer can be built before it.
pnpm install
pnpm run build
pnpm run testReleases are managed via changesets:
pnpm changeset # describe the change
pnpm version-packages # roll versions + regenerate CHANGELOGs
pnpm release-packages # publish to npm