Changelog - 2026-07-08
Typesense Advanced Search
New Feature Breaking ChangeIn one line. The Typesense connector now supports vector, semantic and hybrid search, multi-search across collections, and declarative synonyms, all through one search({ mode }) method.
What changed
- Unified
search({ mode }).modeiskeyword,semantic(vector),hybridorraw(native passthrough). It has one typed, OpenAPI-generated schema. - Vector fields.
field.vector('embedding', ...)in the collection DSL. Givedimensionsanddistancefor client-provided vectors, orembed: { from, model }to have Typesense generate the vector at index time. multiSearch.dataSource.multiSearch({ searches, union?, commonParams?, options? })returns side-by-side results per collection. Passunion: trueto merge them into one ranked set.- Declarative synonyms.
synonyms: [{ id, synonyms, root? }]ondefineSearchCollection, provisioned atconfigure(). Withoutrootevery listed term is interchangeable. Withroot, a query for the root expands to its synonyms. defineSearchController. GeneratesPOST /searchandPOST /multi-searchfrom a repository. Each endpoint can be customized or disabled.- Capabilities.
getCapabilities()now reportssearch: { vector, multi, union, synonyms }. - Extra search parameters. Faceting, highlighting, grouping and result-tuning fields are accepted as optional camelCase fields on every non-raw mode.
- Raw escape hatches.
search({ mode: 'raw', params }),getConnector().multiSearch(...)andgetClient()reach anything the typed API does not model.
typescript
await articleRepository.search({
mode: 'semantic',
vectorField: 'embedding',
queryText: 'wireless earbuds',
});Who is affected
- Typesense keyword search users. No action needed. The new features are opt-in.
- Callers of the raw
search({ params })form from 2026-07-05. Addmode: 'raw'.
Breaking changes
WARNING
search() now requires a mode field. A bare search({ params }) call no longer type-checks.
Before:
typescript
const result = await articleRepository.search({ params: { q: 'typescript', query_by: ['title'] } });After:
typescript
const result = await articleRepository.search({ mode: 'raw', params: { q: 'typescript', query_by: ['title'] } });Details
semanticandhybriddefault toprefix: false, because prefix matching is meaningless for vector search and remote embedders (OpenAI, Google, ...) reject it. You can still setprefix.- Synonyms use the Typesense v30+ global synonym-sets model, one set per collection, linked at
configure(). Manage them at runtime withupsertSynonymSet,getSynonymSet,listSynonymSets,deleteSynonymSetandlinkSynonymSetson the connector. - The framework has no retry, circuit breaker or change-data-capture. Keeping the database and the search index in sync is up to you.
See also
- Unified Repository & Connectors Architecture
- Connectors Consistency Hardening - moved engine-specific tuning fields to
engineParams