Weaviate v1.33 is now available open-source and on Weaviate Cloud. This release brings compression by default for optimal resource utilization, 1-bit rotational quantization (RQ), OIDC group management, server-side batch imports and collection aliases become generally available (GA).
There's also expanded filtering capabilities with new ContainsNone and Not operators, plus a few additional improvements across the board.
Here are the release ⭐️highlights⭐️!
- Collection aliases goes GA
- Compression by default (& 8-bit RQ goes GA)
- 1-bit rotational quantization (RQ)
ContainsNoneandNotfilter operators- OIDC group management
- Server-side batch imports
- Community contributions
Collection aliases goes GA
Collection aliases, introduced as a preview in Weaviate v1.32, are now generally available (GA) and ready for production use. This feature provides an elegant solution for seamless collection migrations and production workflows.
Collection aliases allow you to:
- Create references to collections that can be updated without downtime
- Perform zero-downtime migrations when changing collection settings
- Simplify production deployments with stable collection references
Creating and using aliases:
# Create an alias
client.alias.create(alias_name="ProductsAlias", target_collection="Products_v1")
# Use the alias like a regular collection
products = client.collections.use("ProductsAlias")
results = products.query.near_text("wireless headphones")
# Update the alias to point to a new collection (zero downtime!)
client.alias.update(alias_name="ProductsAlias", new_target_collection="Products_v2")
This GA release includes full client library support and comprehensive documentation, making collection aliases a robust tool for production environments.
Compression by default (& 8-bit RQ goes GA)
Starting with Weaviate v1.33, we're making efficient resource utilization the default experience. 8-bit rotational quantization (RQ) is now enabled by default when creating new collections, ensuring you get optimal memory usage and faster performance out of the box.
This means that new collections will automatically benefit from:
- Up-to 4x memory compression with 98-99% recall maintained
- No training phase required - RQ works immediately at index creation
- No configuration needed - just create a new collection to get started
Additionally, 8-bit RQ is now generally available (GA), meaning it's fully supported and ready for production workloads. Our extensive testing has shown that RQ provides the best balance of compression, performance, and recall for most use cases.
If you need different behavior, you can customize this through the DEFAULT_QUANTIZATION environment variable:
# To disable compression by default
DEFAULT_QUANTIZATION=none
# To use a different quantization method
DEFAULT_QUANTIZATION=pq
1-bit rotational quantization (RQ)
Building on the success of 8-bit RQ, Weaviate v1.33 introduces 1-bit rotational quantization (RQ) as a preview feature. This quantization technique provides massive compression rates while maintaining good search quality.
1-bit RQ works through an asymmetric quantization approach:
- Data vectors are quantized using just 1 bit per dimension (storing only the sign)
- Query vectors use 5 bits per dimension during search for improved accuracy
- Fast pseudorandom rotation distributes information evenly across dimensions
This asymmetric design provides several key advantages:
- Close to 32x compression as dimensionality increases
- More robust than binary quantization (BQ) - works well even on some datasets where BQ performs poorly
Our internal testing shows that 1-bit RQ serves as a good alternative to binary quantization, with only a slight decrease in throughput compared to BQ, but significantly better recall on some datasets.
1-bit RQ was added in v1.33 as a preview feature. This means it's still under development and may change in future releases. We don't recommend using it in production environments at this time.
OIDC group management
Managing permissions for large teams just became much more streamlined with OIDC group management. You can now leverage user groups defined in your identity provider (like Keycloak, Okta, or Auth0) to manage permissions in Weaviate automatically.
Here's how it works:
- Groups are defined in your identity provider
- User group memberships are passed to Weaviate in the OIDC token
- Roles are assigned to these OIDC groups in Weaviate
- Users automatically inherit permissions based on their group memberships
Assign roles to an OIDC group:
# Assign multiple roles to a group
admin_client.groups.oidc.assign_roles(
group_id="/admin-group",
role_names=["testRole", "viewer"]
)
List roles assigned to a group:
# Get all roles for a specific group
group_roles = admin_client.groups.oidc.get_assigned_roles(
group_id="/admin-group",
include_permissions=True
)
print(f"Roles assigned to '/admin-group': {list(group_roles.keys())}")
Revoke roles from a group:
# Remove specific roles from a group
admin_client.groups.oidc.revoke_roles(
group_id="/admin-group",
role_names=["testRole"]
)
This feature eliminates the need to assign roles to each user individually, making it much easier to manage access for large teams and ensuring permissions stay in sync with your organization's structure.
ContainsNone and Not filter operators
Weaviate v1.33 expands your filtering capabilities with two new operators: ContainsNone and Not. These operators provide more flexible and intuitive ways to express complex filtering logic.
ContainsNone Operator
The ContainsNone operator works similarly to existing ContainsAny and ContainsAll operators, but returns objects where the property contains none of the specified values:
# Find articles that contain none of these topics
results = collection.query.fetch_objects(
where=Filter.by_property("topics").contains_none(
["politics", "sports", "entertainment"]
),
return_properties=["title", "content"],
limit=10
)
Internally, ContainsNone is efficiently translated to:
Not( (prop Equals "politics") Or (prop Equals "sports") Or (prop Equals "entertainment") )
Not Operator
The Not operator provides negation of its single operand, allowing you to express inverse conditions clearly:
# Find products that are NOT in the electronics category
results = product_collection.query.fetch_objects(
where=wvc.query.Filter.not_(
Filter.by_property("category").equal("electronics")
),
return_properties=["name", "price"],
limit=10
)
These operators work seamlessly with Weaviate's existing filtering system and can be combined with other operators for sophisticated queries.
Server-side batch imports
Server-side batching was added in v1.33 as a preview feature and is not supported in our client libraries yet.
Data import just got a whole lot easier with server-side batch imports (also called automatic batching). Instead of manually tuning batch parameters on the client side, you can now let the server manage the data flow rate for optimal performance. Through dynamic backpressure the server continuously monitors its internal queue size and calculates an exponential moving average (EMA) of its workload. It then tells the client the ideal number of objects to send in the next chunk.
Using server-side batching in Python is straightforward:
# Server-side batching automatically optimizes the data ingestion flow
with client.batch.automatic() as batch:
for data_object in data_objects:
batch.add_object(
properties=data_object,
collection="Articles"
)
Why use automatic batching?
- Simplified client code - No need to manually tune batch size and concurrent requests
- Improved stability - Automatic backpressure prevents server overloads and timeouts
- Enhanced resilience - Better handling of cluster scaling events and long-running vectorization tasks
Community contributions
Weaviate is an open-source project, and we're always thrilled to see contributions from our amazing community.
For this release, we are super excited to shout-out the following contributor for their contributions to Weaviate. 🎉🎉🎉
- @yazan-abu-obaideh contributed #8811
If you're interested in contributing to Weaviate, please check out our contribution guide, and browse the open issues on GitHub. Look for the good-first-issue label to find great starting points!
Summary
Weaviate v1.33 represents a significant step forward in making vector databases more efficient, easier to use, and more powerful out of the box. With compression enabled by default, advanced quantization options, streamlined batch imports, and enhanced permission management, this release sets you up for success whether you're just getting started or scaling to production.
Ready to get started?
The release is available open-source as always on GitHub, and is already available for new Sandboxes on Weaviate Cloud.
For those upgrading a self-hosted version, please check the migration guide for version-specific notes.
It will be available for Serverless clusters on Weaviate Cloud soon as well.
Thanks for reading, and happy vector searching! 👋
Ready to start building?
Check out the Quickstart tutorial, or build amazing apps with a free trial of Weaviate Cloud (WCD).
Don't want to miss another blog post?
Sign up for our bi-weekly newsletter to stay updated!
By submitting, I agree to the Terms of Service and Privacy Policy.