Changelog¶

Version 3.26.0¶

This minor version includes:

Support for Weaviate 1.23
Bump of the default version for Weaviate Embedded DB to v1.23.0
Adds support for nodes api verbosity option

Version 3.25.3¶

This patch version includes

Bump of the default version for Weaviate Embedded DB to v1.22.3

Version 3.25.2¶

This patch version includes

Fixes to the codebase naming convention and directory structure to prevent collision with Google’s proto-plus library
Fixes to the build method so that readthedocs.io builds the documentation correctly again

Version 3.25.1¶

This patch version includes:

Bump default embedded version to 1.22.0

Version 3.25.0¶

This minor version includes:

Support for new Weaviate nested objects on insert and query
- client.data_object.create() now supports nested objects
- client.query.get() now supports nested objects
Updates to use Weaviate’s v1 gRPC API
Support for batching with Weaviate>1.22.0 version and async vector indexing
Addition of the client.batch.wait_for_async_indexing() method to force block until async indexing is complete
Add tests for Python 3.12 to ensure compatibility

Version 3.24.2¶

This patch version includes:

Small fix to the batching process to ensure that failed multi-tenant objects are re-added to the batch with their tenant attached

Version 3.24.1¶

This patch version updates the changelog.rst that became stale over the last few releases

Version 3.24.0¶

This minor version includes:

Small fixes and improvements throughout the codebase:
- Catching and reraising of JsonDecodeException for users to catch
- Client-wide mypy error fixing and type hinting improvements
- Fix for where filter operands in batch.delete_objects
- Removal of buggy client-side schema validation
- Package dependency updates

Version 3.23.2¶

This patch version includes:

Enforcing class name capitalization throughout the client
Further fixes to where filtering with ContainsAny/All

Version 3.23.1¶

This patch version includes:

Enabling of rerank-cohere module in EmbeddedWeaviate
Fixes for where filtering between query.get over GraphQL and batch.delete_objects over REST

Version 3.23.0¶

This minor version updates the client to work with Weaviate’s 1.21 version and includes:

Adds support for near<Media> filters when using the new multi2vec-bind module for neural searching on different media types
- client.query.get().with_near_audio()
- client.query.get().with_near_depth()
- client.query.get().with_near_image() (unchanged from previous versions but usable by the module)
- client.query.get().with_near_imu()
- client.query.get().with_near_thermal()
- client.query.get().with_near_video()
Deprecates configuring client.batch using client.batch() in favour of using client.batch.configure()
- client.batch() will be removed in a future version
- client.batch.configure() will return None in a future version
- with client.batch as batch should be the standard way to initiate a batch
Adds support for new ContainsAny and ContainsAll filters when using .with_where
Adds support for updating individual tenants within a multi-tenancy class configuration: client.schema.update_class_tenants
Improves client.batch algorithm to choose batch size dynamically maximising throughput
Provides sensible defaults to client.batch that do not cause unexpected damaging consequences like infinite batch sizes
Fixes bugs when using .with_where with valueText, valueString, and valueGeoRange types

Version 3.22.1¶

This patch version includes:

Fix “is client outdated”-check in air-gaped environments
Add tenant to batch delete

Version 3.22.0¶

This minor version includes:

Multi-tenancy
Aggregate with limit
Autocut
Fusion type for hybrid search
Client emits a warning when it is outdated (three minor version behind last release on pypi)
Increase default embedded version to 1.19.12

Version 3.21.0¶

This minor version includes: - Weaviate Embedded supports MacOs

Version 3.20.1¶

This patch version includes:

Fix imports without GRPC package
Improve shutdown handling with Weaviate Embedded

Version 3.20.0¶

This minor version includes:

Increase maximum version of request library to 2.31.0. This also updates to urllib 2.0. This may contain minor breaking changes if you use urllib in other projects in the same virtual environment.
Add licensing information to pypi package
Increase default embedded version to 1.19.7

Version 3.19.2¶

This patch version includes:

Add custom headers to all requests
Support properties field in generative groupedResult field

Version 3.19.1¶

This patch version includes:

Fixes imports of of weaviate_pb2.

Version 3.19.0¶

This minor version includes:

Increases default embedded version to 1.19.3
Clients emits warning if used weaviate version is too old (3 versions behind latest minor version)

Adds native support for querying reference properties

result = client.query.get(
  "Article", ["title", "url", "wordCount", LinkTo(link_on="caller", linked_class="Person", properties=["name"])]
     )

Adds dataclasses to easier access to additional properties

query = client.query.get("Test").with_additional(
            weaviate.AdditionalProperties(
                uuid=True,
                vector=True,
                creationTimeUnix=True,
                lastUpdateTimeUnix=True,
                distance=True,
            )
        )

Typing fixes
Expand support for experimental GRPC API and add support for
- BM25 and hybrid search
- Additional properties (via dataclass shown above)
- Querying reference properties (via dataclass shown above)

Version 3.18.0¶

This minor version includes:

Add support for properties with hybrid search
Fixes documentation publishing on readthedocs

Version 3.17.1¶

This patch version includes:

Fix schemas with new property keys indexFilterable and indexSearchable.

Version 3.17.0¶

This minor version includes:

Add support for groupBy to group objects:

.with_group_by(properties=["caller"], groups=2, objects_per_group=3)

Add support for uuid and uuid[] datatypes.
Add schema.exists(class).

Add support for Support GQL Get{} tunable consistency

resp = (
    client.query.get("Article", ["name"])
    .with_additional("isConsistent")
    .with_consistency_level(ConsistencyLevel.ALL)
    .do()
)

Version 3.16.2¶

This patch version includes:

Fix url containing username and password.

Version 3.16.1¶

This patch version includes:

Fixes timeout error in detection of grpc.

Version 3.16.0¶

This minor version includes:

Experimental support for GRPC.
- Can by enabled by installing the client with pip install weaviate-client[GRPC] or install the grpcio package manually.
- To disable uninstall the grpcio package.
- This will speed up certain GraphQL queries: Get with NearObject or NearVector if only non-reference queries are retrieved and no other options are set.
Removal of python 3.7 support. Minimum supported version is python 3.8
Removal of the WCS module. Note that the module was used to administrate old WCS instances and does not work anymore.

Version 3.15.6¶

This patch version includes:

Fix multi-line queries for BM25 and hybrid search.

Version 3.15.5¶

This patch version includes:

EmbeddedDB now supports latest and versions (eg 1.18.3) as version argument.
Removed cluster_hostname from EmbeddedOptions. It can still be set by using additional_env_vars.
Fix multi-line queries for generative search.

Version 3.15.4¶

This patch version includes:

Fix imports of EmbeddedDB on Mac. It now properly raises an exception that MacOS is currently unsupported.

Version 3.15.3¶

This patch version includes:

Improve embedded weaviate: Better folder structures, add support for env variables and support multiple versions.
Fix edge case for timeout retries: When all objects have been added, no empty batch will be send.
Fix authentication via additional_headers

Version 3.15.2¶

This patch version includes:

Fixes API keys with Weaviate setups that do not have OIDC enabled.

Version 3.15.1¶

This patch version includes:

Fixes refreshing of OIDC tokens on unstable connections

Version 3.15.0¶

This minor version includes:

GraphQL Multiple queries and aliases support

client.query.multi_get(
        [
           client.query.get("Ship", ["name"]).with_alias("one"),
           client.query.get("Ship", ["size"]).with_alias("two"),
           client.query.get("Person", ["name"])
        ]

Adds support for embedded weaviate version

from weaviate import Client
from weaviate.embedded import EmbeddedOptions

# Create the embedded client which automatically launches a Weaviate database in the background
client = Client(embedded_options=EmbeddedOptions())

Version 3.14.0¶

This minor version includes:

Support for API-Keys

client = weaviate.Client(url, auth_client_secret=AuthApiKey(api_key="my-secret-key"))

Version 3.13.0¶

This minor version includes:

Extend CRUD operations for single data objects and reference with consistency level.
Extend batch operations with consistency level.
Add Cursor api.
Add support for azure backup module.

Version 3.12.0¶

This minor version includes:

Adds with_generate in GetBuilder() which allows to use the generative openai module. Needs Weaviate with version >=v1.17.3.
Fix for empty OIDC scopes
New startup_period parameter in Client(). The client will wait for the given timeout for Weaviate to start. By default 5 seconds.
Improved error messages for where filters and authentication.

Version 3.11.0¶

This minor version includes:

New status code attribute for UnexpectedStatusCodeException that can be accessed like this:

try:
    # your code
except weaviate.UnexpectedStatusCodeException as err:
    print(err.status_code)

Fix for get_meta().
Caches server version at Client initialization. This improves batch reference creation performance.
Changes accepted data types for arguments from_object_uuid and to_object_uuid of the method add_reference() to str and uuid.UUID.

Adds automatic retry for failed objects. It can be configured using the weaviate_error_retries argument for the configure() or __call__(), and should be an instance of WeaviateErrorRetryConf. It can be used like this:

All errors:

from weaviate import WeaviateErrorRetryConf

with client.batch(
    weaviate_error_retries=WeaviateErrorRetryConf(number_retries=3),
) as batch:
    # Your code

Exclude errors, all the other errors will be retried:

from weaviate import WeaviateErrorRetryConf

with client.batch(
    weaviate_error_retries=WeaviateErrorRetryConf(number_retries=3, errors_to_exclude=["Ignore me", "other error to ignore"]),
) as batch:
    # Your code

Include errors, all the other errors will be ignored:

from weaviate import WeaviateErrorRetryConf

with client.batch(
    weaviate_error_retries=WeaviateErrorRetryConf(number_retries=3, errors_to_include=["error to retry", "other error to test again"]),
) as batch:
    # Your code

Adds new arguments sort and offset for get().

Version 3.10.0¶

This minor version includes:

Improves error message for error "413: Payload Too Large"

Adds new Client credential OIDC flow method:

client_credentials_config = weaviate.AuthClientCredentials(
    client_secret = "client_secret",
    scope = "scope1 scope2" # optional, depends on the configuration of your identity provider
)
client = weaviate.Client("https://localhost:8080", auth_client_secret=client_credentials_config)

Improves size of batches on dynamic batching.
New limit argument to get() method of the DataObject client attribute.
Bump minimum version of request to 2.28.0
Adds support for node_name and consistency_level for both get() and get_by_id() of the DataObject client attribute. This can be used ONLY with Weaviate Server v1.17.0 or later.
Adds support for replication factor in schema. This can be used ONLY with Weaviate Server v1.17.0 or later. This can be configured in class schema like this:
my_class = { "class": "MyClass", ..., "replicationConfig": { "factor": 1 } }
Adds support for Bm25 for Get queries, with_bm25(). This can be used ONLY with Weaviate Server v1.17.0 or later.
Adds support for with_hybrid for Get queries, with_hybrid(). This can be used ONLY with Weaviate Server v1.17.0 or later.

Version 3.9.0¶

This minor version includes:

Authentication using Bearer token, by adding additional_headers to the Client initialization:

client = weaviate.Client(
    url='http://localhost:8080',
    additional_headers={
        {"authorization": "Bearer <MY_TOKEN>"}
    }
)

Multi-threading Batch import:
- Now it is possible to import data using multi-threading. The number of threads can be set using the new argument num_workers in configure() and __call__(), defaults to 1 ( Use with care to not overload your weaviate instance.).
- New argument connection_error_retries to retry on ConnectionError that can be set in configure() and __call__() or using the property getter/setter: client.batch.connection_error_retries to get the value and client.batch.connection_error_retries = 5 to set the value.
- New method start() to create a BatchExecutor (ThreadExecutor). This method does NOT need to be called if using the Batch in a context manager (with). Also it is idempotent.
- New method shutdown() to shutdown the existing BatchExecutor (ThreadExecutor) to release any resources that it is holding once the batch import is done. This method does NOT need to be called if using the Batch in a context manager (with). Also it is idempotent.
New Client attribute Cluster to check the status of the cluster nodes.
- The method get_nodes_status() returns the status of each node as a list of dictionaries.
  client.cluster.get_nodes_status()
Fix for replace() and update() when using with Weaviate server >=v1.14.0.
New default timeout_config: (10, 60).

Version 3.8.0¶

This minor version includes:

Backup functionalities (Backup):
- create() method to create backups (all/subset of classes).
- get_create_status() method to get the status of the created backup.
- restore() method to restore Weaviate from a backup (all/subset of classes).
- get_restore_status() method to get the status of the restored backup.
New Client attribute: backup to create, restore and get status of the backups. All backup operations MUST be done through Client.backup.
Added return value for add_data_object(), it now returns the UUID of the added object, if one was not set then an UUIDv4 will be generated.

Version 3.7.0¶

This minor version includes:

Adds rolling average (last 5 batches) for batch creation time used by Dynamic Batching method.
Adds ability to use get() without specifying any properties IF Additional Properties (with_additional()) are set before executing the query.
Adds base Weaviate Exception WeaviateBaseError.
Adds ability to set proxies. Can be set at Client initialization by using the new proxies or trust_env arguments.
Batch creates UUIDs (UUIDv4) for all added objects that do not have one at client side (fixes data duplication on Batch retries).
Adds new methods for WCS for instances that have authentication enabled:
- get_users_of_cluster() to get users (emails) for all the users that have access to the created Weaviate instance.
- add_user_to_cluster() to add users (email) to the created Weaviate instance.
- remove_user_from_cluster() to remove user (email) from the created Weaviate instance.

Version 3.6.0¶

This minor version includes:

New function in check_batch_result() used to print errors from batch creation.
New function argument class_name for generate_local_beacon(), used ONLY with Weaviate Server version >= 1.14.0
(defaults to None for backwards compatibility).
check_batch_result() is the default callback function for Batch (configure() and __call__()) (instead of None).
New method argument to_object_class_name for add_reference(), used ONLY with Weaviate Server version >= 1.14.0 (defaults to None for backwards compatibility).
Support for distance in GraphQL filters (only with Weaviate server >= 1.14.0).
For DataObject:
- New method argument class_name for get_by_id(), get(), delete() exists(), used ONLY with Weaviate Server version >= 1.14.0 (defaults to None for backwards compatibility).
- Deprecation Warning if Weaviate Server version >= 1.14.0 and class_name is None OR if Weaviate Server version < 1.14.0 and class_name is NOT None.
For Reference:
- New method arguments from_class_name and to_class_name (to_class_names for update()) for add(), delete(), update(), used ONLY with Weaviate Server version >= 1.14.0 (defaults to None for backwards compatibility).
- Deprecation Warning if Weaviate Server version >= 1.14.0 and class_name is None OR if Weaviate Server version < 1.14.0 and class_name is NOT None.

Version 3.5.1¶

This patch version fixes:

the rerank not being set bug in with_ask().
the bug when using double quotes(”) in question field in with_ask().
the bug where nearText filter checks for objects in moveXXX clause but never sets it.

Version 3.5.0¶

This minor version contains functionality for the new features introduced in Weaviate v1.13.0.

New Batch method delete_objects() to delete all objects that match a particular expression (where filter).
New GetBuilder method with_sort() that allows sorting data on a particular field/s.
New AggregateBuilder method with_near_text() that allows to aggregate data that is matching nearText filter.
New AggregateBuilder method with_near_object() that allows to aggregate data that is matching nearObject filter.
New AggregateBuilder method with_near_vector() that allows to aggregate data that is matching nearVector filter.

Version 3.4.2¶

This patch version fixes another bug in exists().

Version 3.4.1¶

This patch version fixes bug in exists().

Version 3.4.0¶

This minor version fixes the bug in setting the Schema’s invertedIndexConfig field.

New method get_class_shards() to get all shards configuration of a particular class.

New method update_class_shard() to update one/all shard/s configuration of a particular class.

Support for new Property field: tokenization.

Version 3.3.3¶

This patch version fixes the nearImage filter requests.

Version 3.3.2¶

This patch version allows using UUIDs in hex format for DataObject too i.e. UUIDs without hyphens.

Version 3.3.1¶

This patch version allows using UUIDs in hex format too i.e. UUIDs without hyphens.

Version 3.3.0¶

This minor version adds a new with_offset() for the Get queries. This method should be used with the with_limit(). This new feature (introduced in weaviate version 1.8.0) allows to use pagination functionality with the Get queries. The offset represents the start index of the objects to be returned, and the number of objects is specified by the with_limit() method.

For example, to list the first ten results, set limit: 10. Then, to “display the second page of 10”, set offset: 10, limit: 10 and so on. E.g. to show the 9th page of 10 results, set offset: 80, limit: 10 to effectively display results 81-90.

Version 3.2.5¶

This patch fixes the 'Batch' object is not callable error.

Version 3.2.4¶

All class_name and cross-refs dataType are implicitly capitalized. (This functionality is added because if class_name is not capitalized then Weaviate server does it for you, and this was leading to errors where the client and server have different configurations.)

Fixes/updates in Schema class:

This patch fixes the contains() to accept separate class schemas as argument i.e. it does not expect to have only this format: {"classes": [CLASS_1, CLASS_2, ...]}; now it is possible to pass just CLASS_X as well.

Version 3.2.3¶

This patch fixes the with_near_object(). It uses now explicit string literals for id/beacon in nearoOject clauses.

Version 3.2.2¶

This patch adds support for array data types: boolean[], date[].

Version 3.2.1¶

This patch adds support for array data types: int[], number[], text[], string[].

Version 3.2.0¶

Fixes/updates in WCS class:

Fixed progress bar for create(), it is being updated in Notebooks too, instead of printing each iteration on new line.
Method create() now prints the creation status above the bar.

Updates in gql sub-package:

New key-value autocorrect: <bool> introduced for the NearText and Ask filters. The autocorrect is enabled only if Weaviate server has the text-spellcheck module enabled. If autocorrect is True the query is corrected before the query is made. Usage example:

# with 'nearText' filter
client.query\
    .get('Article', ['title', 'author'])\
    .near_text(
        {
            'concepts': ['Ecconomy'],
            'autocorrect': True
        }
    )
    # the concept should be corrected to 'Economy'
# with 'ask' filter
client.query\
    .get('Article', ['title', 'author'])\
    .with_ask(
        {
            'question': 'When was the last financial crysis?',
            'autocorrect': True
        }
    )
    # the question should be corrected to 'When was the last financial crisis?'

New method with_additional() is added to GET the _additional properties. Usage example:

# single additional property with this GraphQL query
'''
{
    Get {
        Article {
            title
            author
            _additional {
                id
            }
        }
    }
}
'''
client.query\
    .get('Article', ['title', 'author'])\
    .with_additional('id') # argument as `str`

# multiple additional property with this GraphQL query
'''
{
    Get {
        Article {
            title
            author
            _additional {
                id
                certainty
            }
        }
    }
}
'''
client.query\
    .get('Article', ['title', 'author'])\
    .with_additional(['id', 'certainty']) # argument as `List[str]`

# additional properties as clause with this GraphQL query
'''
{
    Get {
        Article {
            title
            author
            _additional {
                classification {
                    basedOn
                    classifiedFields
                    completed
                    id
                    scope
                }
            }
        }
    }
}
'''
client.query\
    .get('Article', ['title', 'author'])\
    .with_additional(
        {
            'classification' : [
                'basedOn',
                'classifiedFields',
                'completed',
                'id',
                'scope'
            ]
        }
    ) # argument as `Dict[str, List[str]]`

# or with this GraphQL query
'''
{
    Get {
        Article {
            title
            author
            _additional {
                classification {
                    completed
                }
            }
        }
    }
}
'''
client.query\
    .get('Article', ['title', 'author'])\
    .with_additional(
        {
            'classification' : 'completed'
        }
    ) # argument as `Dict[str, str]`

# additional properties as clause and clause settings with this GraphQL query
'''
{
    Get {
        Article {
            title
            author
            _additional {
                token (
                    properties: ["content"]
                    limit: 10
                    certainty: 0.8
                ) {
                    certainty
                    endPosition
                    entity
                    property
                    startPosition
                    word
                }
            }
        }
    }
}
'''
clause = {
    'token': [
        'certainty',
        'endPosition',
        'entity',
        'property',
        'startPosition',
        'word',
    ]
}
settings = {
    'properties': ["content"],  # is required
    'limit': 10,                # optional, int
    'certainty': 0.8            # optional, float
}
client.query\
    .get('Article', ['title', 'author'])\
    .with_additional(
        (clause, settings)
    ) # argument as `Tuple[Dict[str, List[str]], Dict[str, Any]]`

# if the desired clause does not match any example above, then the clause can always
# be converted to string before passing it to the `.with_additional` method

Version 3.1.1¶

Fixes in WCS class:
- Make WCS’s methods’ argument cluster_name case insensitive (lowercased inside the method) to match Weaviate Cloud Service’ naming convention, this fixes the error when Weaviate Cloud Service lowercases the cluster_name but the users are not aware of this and get the exception KeyError.

Version 3.1.0¶

New Batch methods:
- pop_object() / pop_reference() to remove and return an added object/reference from the Batch at position index (by default -1).
- empty_objects() / empty_references() to remove all the existing objects/references from the Batch instance.
- is_empty_objects() / is_empty_references() to check there are any objects/references in the Batch instance.
Fixes in WCS class:
- Authentication only with AuthClientPassword.
- The create() argument module is renamed to modules and can also be a list of modules to enable for the WCS cluster. The argument can be used on the PROD WCS too.
- The get_cluster_config() does not raise an exception if the cluster does not exist but returns a empty configuration.
- The delete_cluster() does not raise an exception if the cluster does not exist.
Add phoneNumber to the Weaviate’s primitive types. Thanks to GitHub user @cdpierse.
Bug fix in Connection.
Fix ConnectionError handling.
Optimization in weaviate.batch.requests and weaviate.connect.connection.

Version 3.0.0¶

weaviate.tools module is REMOVED.
- Batcher class is REMOVED.
- WCS class is moved from the weaviate.tools to the new module weaviate.wcs
- weaviate.tools.generate_uuid is REMOVED.
weaviate.util.generate_uuid5() is ADDED.
New Batch class implementation to replace the old one. This implementation uses the BatchRequest objects under the hood, which means that there is no need to create BatchRequest’s anymore. This new class implementation allows 3 different batch creations methods: manual, auto-create and auto-create with dynamic batching. See the Batch documentation for more information.
BatchRequest classes (ObjectsBatchRequest and ReferenceBatchRequest) are hidden from the user and should not be used anymore. This is due to the new Batch class implementation.
New Schema field is ADDED, “shardingConfig”. It can bu used with Weaviate version >= 1.6.0.
New method update_config() used to update mutable schema configuration (like efConstruction, …).