Hybrid

Milvus implementation of hybrid search capability.

This module provides a Milvus implementation of the HybridCapability protocol, combining BM25-backed sparse retrieval with dense vector similarity search. Encryption and batching are handled by the base class.

MilvusHybridCapability(collection_name, client, uri, config, query_field=CHUNK_KEYS.CONTENT, index_type=DEFAULT_INDEX_TYPE, index_params=None, default_index_params_map=None, distance_metric=DEFAULT_DISTANCE_METRIC, id_max_length=100, content_max_length=65535, encryption=None, default_batch_size=None)

Bases: BaseHybridCapability

Milvus hybrid capability backed by BM25 sparse search and dense vector search.

Attributes:

Name	Type	Description
collection_name	str	Name of the target Milvus collection.
client	AsyncMilvusClient	Async Milvus client used for collection operations.

Initialize the Milvus hybrid capability.

Parameters:

Name	Type	Description	Default
collection_name	str	Name of the Milvus collection.	required
client	AsyncMilvusClient	Async Milvus client instance.	required
uri	str	Milvus URI used to create the sync client for index params.	required
config	list[SearchConfig]	Hybrid search configuration entries.	required
query_field	str	Field used to store chunk content. Defaults to content.	CONTENT
index_type	str	Dense index type for vector fields. Defaults to "IVF_FLAT".	DEFAULT_INDEX_TYPE
index_params	dict[str, Any] | None	Dense index-specific parameters. Defaults to None.	None
default_index_params_map	dict[str, dict[str, Any]] | None	Default index parameters keyed by Milvus index type when index_params is not provided. Defaults to None.	None
distance_metric	str	Dense vector metric type. Defaults to "IP".	DEFAULT_DISTANCE_METRIC
id_max_length	int	Maximum length for chunk IDs. Defaults to 100.	100
content_max_length	int	Maximum length for content fields. Defaults to 65535.	65535
encryption	EncryptionCapability | None	Encryption capability. Defaults to None.	None
default_batch_size	int | None	Default batch size for batched operations. Defaults to None.	None

Raises:

Type	Description
ValueError	If no search configuration is provided.

fulltext_configs property

Return configured fulltext searches.

Returns:

Type	Description
list[SearchConfig]	list[SearchConfig]: Fulltext search configurations.

vector_configs property

Return configured dense vector searches.

Returns:

Type	Description
list[SearchConfig]	list[SearchConfig]: Dense vector search configurations.

create(chunks, batch_size=None, **kwargs) async

Create chunks, enriching them with dense vectors before persistence.

Parameters:

Name	Type	Description	Default
chunks	list[Chunk]	Chunks to persist.	required
batch_size	int | None	Batch size override. Defaults to None.	None
**kwargs	Any	Backend-specific arguments forwarded to upsert.	{}

ensure_index() async

Ensure the hybrid collection exists with BM25 and dense indexes.

Raises:

Type	Description
ValueError	If a required vector dimension cannot be inferred.

update(update_values, filters=None, batch_size=None, **kwargs) async

Update chunks and regenerate dense vectors when text changes.

Parameters:

Name	Type	Description	Default
update_values	dict[str, Any]	Values to update.	required
filters	FilterClause | QueryFilter | None	Filters selecting the target rows. Defaults to None.	None
batch_size	int | None	Batch size override. Defaults to None.	None
**kwargs	Any	Backend-specific arguments forwarded to upsert.	{}