< Docs
  • Python


This method conducts a vector similarity search.


search(data, anns_field, param, limit, expr=None, partition_names=None, output_fields=None, timeout=None, round_decimal=-1, **kwargs)


dataData to search withlist[list[Float]]True
anns_fieldName of the vector field to search onStringTrue
paramSpecific search parameter(s) of the index on the vector field. For details, refer to Prepare search parameters.DictTrue
limitNumber of nearest records to return. The sum of this value and offset should be less than 16384.IntegerTrue
exprBoolean expression to filter the dataStringFalse
partition_namesList of names of the partitions to search on.
All partition will be searched if it is left empty.
output_fieldsList of names of fields to output.
When specified, you can get the values of the specified fields by using hit.entity.get().
timeoutAn optional duration of time in seconds to allow for the RPC. If it is set to None, the client keeps waiting until the server responds or error occurs.FloatFalse
round_decimalNumber of the decimal places of the returned distanceIntegerFalse
kwargs: _asyncBoolean value to indicate if to invoke asynchronously.BoolFalse
kwargs: _callbackFunction that will be invoked after server responds successfully. It takes effect only if _async is set to True.FunctionFalse
kwargs: consistency_levelConsistency level used in the search.String/IntegerFalse
kwargs: guarantee_timestampMilvus searches on the data view before this timestamp when it is provided. Otherwise, it searches the most updated data view. It can be only used in Customized level of consistency.IntegerFalse
kwargs: graceful_timePyMilvus will use current timestamp minus the graceful_time as the guarantee_timestamp for search. It can be only used in Bounded level of consistency.IntegerFalse
kwargs: travel_timestampTimestamp that is used for Time Travel. Users can specify a timestamp in a search to get results based on a data view at a specified point in time.IntegerFalse


A SearchResult object, an iterable, 2d-array-like class whose first dimension is the number of vectors to query (nq), and the second dimension is the number of limit (topk).


  • RpcError: error if gRPC encounter an error.
  • ParamError: error if the parameters are invalid.
  • DataTypeNotMatchException: error if wrong type of data is passed to server.
  • BaseException: error if the return result from server is not ok.


search_params = {"metric_type": "L2", "params": {"nprobe": 10}}
from pymilvus import Collection
collection = Collection("book")      # Get an existing collection.
result =
    data=[[0.1, 0.2]], 
    # demontrates the ways to reference a dynamic field.
    expr='$meta["dynamic_field_1"] > 10 and dynamic_field_2 == 10',
    # sets the names of the fields you want to retrieve from the search result.
    output_fields=['title', 'dynamic_field_1', 'dynamic_field_2'], 

for hits in result:
    # get the IDs of all returned hits

    # get the distances to the query vector from all returned hits
    for hit in hits:
        # get the value of an output field specified in the search request.
        # dynamic fields are supported, but vector fields are not supported yet.