ClearML v3.25 introduces native support for vector databases within the Hyper-Datasets feature. This release enables users to store and search embeddings directly inside ClearML, opening the door to powerful custom RAG pipelines. In addition, v3.25 includes expanded orchestration metrics, new Application Gateway UI, and a range of UI upgrades to streamline day-to-day operations.
Native Vector Database Support: Build Custom RAG Workflows with Your Own Data
ClearML now includes native support for vector fields in Hyper-Datasets, unlocking a foundational component for RAG-based applications. Users can store high-dimensional embeddings – whether from text, images, or any other modality – directly within each frame of a Hyper-Dataset.
With integrated vector similarity search in the UI, you can query large datasets semantically instead of relying on traditional metadata or file names. This makes it easy to:
- Store and inspect custom embeddings
- Run live similarity searches on large, unstructured datasets
- Explore and debug RAG logic using your own domain-specific data
- Prototype and iterate on retrieval workflows without leaving the platform
Whether you’re building semantic search over documentation, powering LLM-assisted labeling tools, or developing RAG pipelines, ClearML now provides the vector-native infrastructure you need.
Application Gateway: Visibility and Control from the UI
The Application Gateway now gets a full management interface within the ClearML UI. Platform administrators can:
- View and inspect all deployed application gateways
- Access router configuration details directly
- Run connectivity diagnostics to validate network health
These improvements simplify the management of distributed resources and make it easier to troubleshoot integration issues across provisioned environments.
Quality-of-Life Improvements
Alongside the headline features, v3.25 includes several smaller upgrades that make the platform smoother and more efficient to use:
- Per-GPU metrics in the orchestration dashboard for finer visibility into multi-GPU nodes
- A new global search bar accessible across all UI pages
- Filtering options added to the Model Endpoints table for faster access to production deployments
- A running instance indicator in the Application section to quickly identify active apps
ClearML v3.25 continues our focus on combining infrastructure visibility, ML observability, and LLM workflow enhancements, all under one platform. If you’re ready to take your workflows to the next level, or just want to make your AI stack more efficient, contact us to learn more about ClearML.