fix: use multi_search for long vector queries

Author: haydenhoang

langchain_typesense/vectorstores.py

@@ -26,8 +26,9 @@
     DeleteQueryParameters,
     DocumentSchema,
     DocumentWriteParameters,
-    SearchParameters,
+    MultiSearchParameters,
 )
+from typesense.types.multi_search import MultiSearchRequestSchema
 
 from langchain_typesense._codec import (
     build_vector_query,
@@ -239,7 +240,7 @@ def _build_search_parameters(
         ef: int | None,
         flat_search_cutoff: int | None,
         include_vectors: bool,
-    ) -> SearchParameters:
+    ) -> MultiSearchParameters:
         """Build a vector search request while protecting managed parameters."""
         self._validate_k(k)
         vector_query = self._build_vector_query(
@@ -281,7 +282,8 @@ def _build_search_parameters(
             filter_by = self._to_filter_by(filter)
             if filter_by:
                 parameters["filter_by"] = filter_by
-        return cast(SearchParameters, parameters)
+        parameters["collection"] = self._collection_name
+        return cast(MultiSearchParameters, parameters)
 
     def _build_vector_query(
         self,
@@ -352,7 +354,7 @@ def _build_hybrid_search_parameters(
         distance_threshold: float | None,
         ef: int | None,
         flat_search_cutoff: int | None,
-    ) -> SearchParameters:
+    ) -> MultiSearchParameters:
         """Build a Typesense request that fuses keyword and vector ranking."""
         parameters: dict[str, Any] = {
             "q": query,
@@ -376,7 +378,8 @@ def _build_hybrid_search_parameters(
             filter_by = self._to_filter_by(filter)
             if filter_by:
                 parameters["filter_by"] = filter_by
-        return cast(SearchParameters, parameters)
+        parameters["collection"] = self._collection_name
+        return cast(MultiSearchParameters, parameters)
 
     def _parse_search_response(
         self,
@@ -405,6 +408,49 @@ def _parse_hybrid_search_response(
             metadata_key=self._metadata_key,
         )
 
+    def _build_multi_search_request(
+        self, parameters: MultiSearchParameters
+    ) -> MultiSearchRequestSchema:
+        """Wrap one collection search for the Typesense multi-search endpoint."""
+        return {"searches": [parameters]}
+
+    @staticmethod
+    def _extract_multi_search_result(response: object) -> Mapping[str, Any]:
+        """Extract the single result returned for a one-query multi-search."""
+        if not isinstance(response, Mapping):
+            raise TypesenseVectorStoreError(
+                "Typesense multi-search response must be an object with one result."
+            )
+        results = response.get("results")
+        if (
+            not isinstance(results, list)
+            or len(results) != 1
+            or not isinstance(results[0], Mapping)
+        ):
+            raise TypesenseVectorStoreError(
+                "Typesense multi-search response must contain exactly one result."
+            )
+        result = results[0]
+        if result.get("code") == 404:
+            raise ObjectNotFound(str(result.get("error", "Collection not found.")))
+        return result
+
+    def _perform_search(self, parameters: MultiSearchParameters) -> Mapping[str, Any]:
+        """Run one search through Typesense's POST-based multi-search endpoint."""
+        response = self._require_sync_client().multi_search.perform(
+            self._build_multi_search_request(parameters)
+        )
+        return self._extract_multi_search_result(response)
+
+    async def _aperform_search(self, parameters: MultiSearchParameters) -> Mapping[str, Any]:
+        """Run one search through the native async multi-search endpoint."""
+        if self._async_client is None:  # pragma: no cover - guarded by callers
+            raise RuntimeError("An asynchronous Typesense client is required.")
+        response = await self._async_client.multi_search.perform(
+            self._build_multi_search_request(parameters)
+        )
+        return self._extract_multi_search_result(response)
+
     # ------------------------------------------------------------------
     # Collection management
     # ------------------------------------------------------------------
@@ -817,11 +863,7 @@ def _search_by_vector(
             flat_search_cutoff=flat_search_cutoff,
             include_vectors=include_vectors,
         )
-        response = (
-            self._require_sync_client()
-            .collections[self._collection_name]
-            .documents.search(parameters)
-        )
+        response = self._perform_search(parameters)
         return self._parse_search_response(response, include_vectors=include_vectors)
 
     async def _asearch_by_vector(
@@ -863,9 +905,7 @@ async def _asearch_by_vector(
             flat_search_cutoff=flat_search_cutoff,
             include_vectors=include_vectors,
         )
-        response = await self._async_client.collections[self._collection_name].documents.search(
-            parameters
-        )
+        response = await self._aperform_search(parameters)
         return self._parse_search_response(response, include_vectors=include_vectors)
 
     def similarity_search(
@@ -1257,11 +1297,7 @@ def hybrid_search_with_score(
             ef=ef,
             flat_search_cutoff=flat_search_cutoff,
         )
-        response = (
-            self._require_sync_client()
-            .collections[self._collection_name]
-            .documents.search(parameters)
-        )
+        response = self._perform_search(parameters)
         return self._parse_hybrid_search_response(response)
 
     async def ahybrid_search(
@@ -1352,9 +1388,7 @@ async def ahybrid_search_with_score(
             ef=ef,
             flat_search_cutoff=flat_search_cutoff,
         )
-        response = await self._async_client.collections[self._collection_name].documents.search(
-            parameters
-        )
+        response = await self._aperform_search(parameters)
         return self._parse_hybrid_search_response(response)
 
     # ------------------------------------------------------------------

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "langchain-typesense"
-version = "0.1.0"
+version = "0.1.1"
 description = "LangChain vector store integration for Typesense"
 readme = "README.md"
 requires-python = ">=3.10"

tests/integration_tests/test_vectorstores.py

@@ -60,6 +60,18 @@ def embed_query(self, text: str) -> list[float]:
         return list(self._vectors[text])
 
 
+class LongVectorEmbeddings(Embeddings):
+    """Embedding fixture whose serialized vector exceeds the GET query limit."""
+
+    dimension = 2000
+
+    def embed_documents(self, texts: list[str]) -> list[list[float]]:
+        return [self.embed_query(text) for text in texts]
+
+    def embed_query(self, text: str) -> list[float]:
+        return [1.0] + [0.0] * (self.dimension - 1)
+
+
 def client_config() -> ConfigDict:
     """Return the local integration-test Typesense configuration."""
     return {
@@ -182,6 +194,23 @@ def test_sync_add_search_score_vector_mmr_and_delete(self, store: TypesenseVecto
         assert store.delete(delete_all_documents=True) is True
         assert store.similarity_search("alpha") == []
 
+    def test_large_vector_query_uses_multi_search(self) -> None:
+        config = client_config()
+        collection_name = f"langchain-typesense-long-query-{uuid4()}"
+        store = TypesenseVectorStore(
+            client=typesense.Client(config),
+            embedding=LongVectorEmbeddings(),
+            collection_name=collection_name,
+        )
+        try:
+            store.add_texts(["large vector"], ids=["large"])
+            assert [document.id for document in store.similarity_search("large vector", k=1)] == [
+                "large"
+            ]
+        finally:
+            store.delete_collection()
+            store.close()
+
     async def test_native_async_add_search_mmr_get_and_delete(
         self, store: TypesenseVectorStore
     ) -> None: