Skip to main content

​
Parameters

ParameterTypeDefaultDescription
configConfigRequiredA validated and immutable Config object containing all necessary parameters for the provider’s operation

​
Functions

​
connect

Establishes a connection to the Milvus database. This method uses the connection parameters from self._config.core to initialize the database client and verify the connection. It also pre-loads a handle to the collection if it already exists. Raises:
  • VectorDBConnectionError: If the connection fails.

​
disconnect

Gracefully terminates the connection to the Milvus database. Resets the internal state of the provider.

​
is_ready

Performs a health check to ensure the database is responsive. Returns:
  • bool: True if the database is connected and responsive, False otherwise.

​
create_collection

Creates the collection and all specified indexes based on the full config. This method dynamically builds the schema and then proceeds to create up to three different types of indexes (dense, sparse, and full-text) based on the flags set in the IndexingConfig. Raises:
  • VectorDBConnectionError: If not connected to Milvus.
  • VectorDBError: If the collection creation fails.

​
delete_collection

Permanently deletes the collection specified in the config. Raises:
  • VectorDBConnectionError: If not connected to the database.
  • CollectionDoesNotExistError: If the collection to be deleted does not exist.
  • VectorDBError: For other Milvus-related errors.

​
collection_exists

Checks if the collection specified in the config already exists. Returns:
  • bool: True if the collection exists, False otherwise.
Raises:
  • VectorDBConnectionError: If not connected to the database.
  • VectorDBError: If checking collection existence fails.

​
upsert

Adds or updates data, now with support for sparse vectors. This method validates the presence of required vector types based on the collection’s schema, prepares the data in a row-based format (list of dicts), and uploads it to Milvus in batches. Parameters:
  • vectors (List[List[float]]): A list of dense vector embeddings. Can be None if the collection is sparse-only.
  • payloads (List[Dict[str, Any]]): A list of corresponding metadata objects.
  • ids (List[Union[str, int]]): A list of unique identifiers for each data point.
  • chunks (Optional[List[str]]): A list of text chunks.
  • **kwargs: Must contain sparse_vectors: List[Dict[int, float]] if the collection was created with create_sparse_index=True.
Raises:
  • UpsertError: If data ingestion fails, or if the required vector types are missing or have mismatched lengths.

​
delete

Removes data from the collection by their unique identifiers. Parameters:
  • ids (List[Union[str, int]]): A list of specific IDs to remove.
  • **kwargs: Provider-specific options.
Raises:
  • VectorDBError: If the deletion fails.

​
fetch

Retrieves full records (payload and vector) by their IDs. Parameters:
  • ids (List[Union[str, int]]): A list of IDs to retrieve the full records for.
  • **kwargs: Provider-specific options.
Returns:
  • List[VectorSearchResult]: A list of VectorSearchResult objects containing the fetched data.
Raises:
  • VectorDBError: If the fetch operation fails.
Master search dispatcher. Routes queries to the appropriate specialized search method based on provided arguments and collection configuration. Parameters:
  • top_k (Optional[int]): The number of results to return. If None, falls back to the default in the Config.
  • query_vector (Optional[List[float]]): The vector for dense or hybrid search.
  • query_text (Optional[str]): The text for full-text or hybrid search.
  • filter (Optional[Dict[str, Any]]): An optional metadata filter.
  • alpha (Optional[float]): The weighting factor for hybrid search. If None, falls back to the default in the Config.
  • fusion_method (Optional[Literal[‘rrf’, ‘weighted’]]): The algorithm to use for hybrid search (‘rrf’ or ‘weighted’).
  • similarity_threshold (Optional[float]): The minimum similarity score for results. If None, falls back to the default in the Config.
  • **kwargs: Additional provider-specific options.
Returns:
  • List[VectorSearchResult]: A list of VectorSearchResult objects.
Raises:
  • ConfigurationError: If the requested search is disabled or the wrong combination of arguments is provided.
  • SearchError: If any underlying search operation fails.
Performs a pure dense vector similarity search. Parameters:
  • query_vector (List[float]): The vector embedding to search for.
  • top_k (int): The number of top results to return.
  • filter (Optional[Dict[str, Any]]): A metadata filter to apply. Defaults to None.
  • similarity_threshold (Optional[float]): The minimum similarity score for results. Defaults to None.
  • **kwargs: Additional provider-specific options.
Returns:
  • List[VectorSearchResult]: A list of the most similar results.
Raises:
  • ConfigurationError: If dense search is not possible; no dense index was created for this collection.
  • SearchError: If the search operation fails.

​
full_text_search

Performs true full-text search using the best available index. Parameters:
  • query_text (str): The text string to search for.
  • top_k (int): The number of top results to return.
  • filter (Optional[Dict[str, Any]]): A metadata filter to apply. Defaults to None.
  • similarity_threshold (Optional[float]): The minimum similarity score for results. Defaults to None.
  • **kwargs: Additional provider-specific options.
Returns:
  • List[VectorSearchResult]: A list of matching results.
Raises:
  • SearchError: If the search operation fails.
  • ConfigurationError: If the required search field is not provided in kwargs.
Performs native hybrid search using Milvus’s multi-vector search and server-side reranking. Parameters:
  • query_vector (List[float]): The dense vector for the semantic part of the search.
  • query_text (str): The raw text for the keyword/sparse part of the search.
  • top_k (int): The number of final results to return.
  • filter (Optional[Dict[str, Any]]): An optional metadata filter.
  • alpha (Optional[float]): The weight for combining scores. If None, falls back to the default in the Config.
  • fusion_method (Optional[Literal[‘rrf’, ‘weighted’]]): The algorithm to use for fusing results (‘rrf’ or ‘weighted’).
  • similarity_threshold (Optional[float]): The minimum similarity score for results. If None, falls back to the default in the Config.
  • **kwargs: Additional provider-specific options.
Returns:
  • List[VectorSearchResult]: A list of VectorSearchResult objects, ordered by the combined hybrid score.
Raises:
  • ConfigurationError: If hybrid search is only possible on collections with both dense and sparse indexes.
  • SearchError: If hybrid search requires query_sparse_vector in kwargs.
  • SearchError: If the search operation fails.
⌘I