Author: Alice Cima, Kedro community
Kedro 1.3.0 through 1.6.0, Kedro Viz 12.4.0, Kedro Datasets 9.6.0, and Kedro skills 0.2.0 introduce new capabilities for serving pipelines, validating datasets and parameters, supporting vector stores, and guiding AI coding assistants.
The Kedro community has shipped a new wave of releases: Kedro core moves to 1.6.0, alongside Kedro-Viz 12.4.0, Kedro-Datasets 9.6.0, and a new Kedro-skills repository that brings Kedro-specific guidance to AI coding assistants. Beyond new datasets, optional parameter validation, dataset validation, and wider Python support, this round introduces the first building blocks for running Kedro pipelines as long-lived, callable services.
What we’re most excited About: Kedro as a Service and Data Validation
Kedro as a Service
For years, one of the most requested capabilities from the community has been the ability to run a Kedro pipeline repeatedly inside a long-running process and call it programmatically from other systems, rather than re-invoking kedro run for every execution. That need has grown as more teams put Kedro at the center of agentic and generative AI systems that call pipelines on demand.
This release ships the first pieces of that vision. At its core is KedroServiceSession, an early iteration of a session implementation that supports multiple runs and data injection, paired with a new HTTP server layer built on top of it. Running kedro server start spins up a server exposing three endpoints:
- /run – execute the pipeline, optionally passing runtime parameters
- /health – check the server’s health status
- /snapshot – retrieve the current project snapshot
An accompanying inspection API also makes the project snapshot available directly, independent of the server, so tooling can introspect a project’s structure without a full run.
This is early, actively developed work. The team encourages the community to try it out and share feedback. For the full story on why this matters and where it is headed, read Kedro as a Service on the Kedro blog.
Data Validation
Data validation helps ensure data quality by checking that datasets conform to expected schemas, types, ranges, business rules, and other properties. In Kedro, validation can catch data quality issues early in the pipeline, before invalid data reaches downstream nodes or is persisted.
Pandera is provided as the built-in validation adapter, while custom validator classes and functions can be used to integrate other validation approaches.
Key highlights
Kedro 1.3.0, 1.4.0, 1.5.0 and 1.6.0
Major features and improvements across the three releases include:
- Added dataset validation: datasets can declare a validator in their catalog entry, and the DataCatalog enforces it on every load and save
- Added –runner-paramsto kedro run, allowing runner constructor keyword arguments such as max_workers to be passed from the CLI
- Added the node function name to project inspection snapshots as func_name.
- Added node source location metadata (source) to inspection snapshots for displaying node code.
- Added optional runtime_params to get_project_snapshot()
- Optional parameter validation that uses type hints on parameter inputs to auto-validate and instantiate Pydantic models or dataclasses, with no impact on untyped parameters
- The new HTTP server layer, powered by KedroServiceSession, to execute pipelines from HTTP requests through /run, /health, and /snapshot
- The first iteration of KedroServiceSession, a new session implementation that supports multiple runs and data injection
- An inspection API to retrieve a project’s snapshot
- A new CLI command, kedro server start, to launch the server
- Official support for Python 3.14
Documentation. The Amazon EMR Serverless, AWS Step Functions, and AWS Batch deployment guides were revamped with a strategy-first layout and more end-to-end detail: custom container images and Spark configuration for EMR, namespace grouping and a full Spaceflights walkthrough for Step Functions, and a custom AWSBatchRunner with S3-backed catalog configuration.
Kedro Datasets 9.6.0
- Added AbstractVectorStoreDataset and vectorstore_base.VectorStoreHandle, backend-agnostic abstract base classes for building vector store datasets
- Three new experimental datasets:
- WeaviateVectorStoreDataset: loads a handle for adding, searching, and deleting entries in Weaviate vector database collections
- FAISSVectorStoreDataset: loads a handle for adding, searching, and deleting entries in FAISS vector indexes
- FeastDataset: handles storing and retrieving features from Feast
Kedro Viz 12.4.0
- Added support for Python 3.14
- Exposed the full internal Kedro node name as kedro_node_name
- Added NodeContextMenu support for VS Code
- Added a Consent Management Platform script to the Kedro-Viz documentation
Kedro skills 0.2.0
Kedro-skills is a PyPI package that distributes AI coding skills to Kedro projects. One command installs contextual guidance that activates automatically when an AI assistant edits matching files.
- kedro skills install catalog-config: installs the catalog configuration skill with the full file layout across all supported IDEs
- kedro skills list: shows available skills and their install status
- kedro skills update: re-renders all installed skills after a package upgrade
- kedro skills uninstall <id>: clean removal of all managed files
- Supported IDEs include Cursor, GitHub Copilot, Claude Code, and Codex CLI
- The included catalog-config skill provides Kedro data catalog configuration guidance for conf/**/*.yml files
From the community
Alongside the core release, the community has been publishing its own explorations of what Kedro can do:
- Building a feature store with Kedro and Feast looks at how Every Cure combines Kedro and Feast to build a reusable feature platform for biomedical machine learning. The same engineered features power both batch analytics and real-time applications.
- Building biomedical knowledge for agentic systems shows how OptimusKG uses Kedro to build a reproducible biomedical knowledge graph for AI agents. It is a clear case for why data infrastructure is a core component of modern agentic systems.
If you are building something interesting with Kedro, the team would love to hear about it. Reach out on Slack or open a pull request against awesome-Kedro.
New use cases
There are also two new hands-on examples in the Kedro Academy:
- kedro-healthcare-graphrag is a walkthrough of building production-ready GraphRAG systems with Kedro. A GraphRAG system maintains three representations of the same underlying data: a knowledge graph for relationship traversal, a vector index for semantic retrieval, and a relational store for aggregate statistics. These need to stay consistent with each other as source data changes.
- kedro-reflection-agent is an example reflection agentic workflow built on Kedro pipelines. Each stage of the improvement lifecycle is implemented as a pipeline with clearly defined inputs and outputs, turning agentic reflection into a repeatable, governed workflow.
What is next
KedroServiceSession and the HTTP server are the beginning of a bigger push to make Kedro pipelines easy to run as reusable, callable services, with further iteration planned based on community feedback.
Data validation is next on the roadmap, with a release coming soon.
With Kedro-skills now available, the team plans to keep growing the library to make building with Kedro smoother for users, and would love the community’s help shaping what comes next, whether that is suggesting, improving, or contributing new skills.
On the visualization side, the team is working toward a lighter, more modular Kedro-Viz, where users can pick and use just the parts of the visualization they need without installing the entire Kedro-Viz application.
Now is a good time to get involved and help shape where this goes. Whether you are interested in running pipelines as services, building generative AI and agentic workflows, exploring data validation or data lineage, or improving the developer experience through skills, contributions, ideas, and feedback from the community are always welcome.
Visit the Kedro website, explore the project on GitHub, or join the community on Kedro Slack to get started.
