ZenML Release Notes

Fixed connectivity issues when using GCP service connectors with private GKE clusters that have DNS-based control plane endpoints.

ZenML now properly connects to these clusters using the same DNS endpoint method as gcloud, ensuring seamless access to private Kubernetes environments.

Resolved an issue where Docker build checksum computation would fail when no container engine was available, even with skip_build=True set in your DockerSettings.

The build step now respects the skip_build flag, so pipelines can run in environments where Docker isn't installed.

Organization admins can now invite existing ZenML users directly to their organization when SSO is enabled, bypassing the traditional email invitation flow. Members with existing accounts are added immediately with proper roles and permissions, making team onboarding faster and more seamless.

Fixed several UI issues including tag overflow in columns, improved error messaging for step logs when permissions are missing, corrected artifact version status display, and resolved organization access permission handling. Additionally fixed a bug where mixed-case email addresses could block invitation acceptance.

You can now stream custom events in real time from running pipelines. Call zenml.streaming.publish() from inside any step or dynamic pipeline to push events that consumers can read via Server-Sent Events (SSE). Enable it by setting stream_broker_implementation_source in your server configuration; the initial implementation includes a Redis-based broker with automatic catch-up, gap signaling, and idle cleanup.

A new POST /api/v1/runs/statistics endpoint lets you query aggregated metrics across your pipeline runs. Group by status, pipeline, stack, user, time buckets (hour/day/week/month), metadata values, and tags, and calculate averages, sums, and min/max over run duration, step counts, cached steps, output artifacts, or custom numeric metadata

ideal for building dashboards and analytics.

The Weights & Biases experiment tracker now automatically attaches ZenML pipeline and step metadata to your W&B runs, groups runs by pipeline execution, and records W&B identifiers back into ZenML step metadata. You can also configure custom groups, job types, run configs, explicit or deterministic run IDs, resume behavior, and pass through additional wandb.init keyword arguments.

The ZenML server now supports structured logging with OpenTelemetry instrumentation. Configure console output with the new ZENML_CONSOLE_LOGGING_FORMAT environment variable, choosing between console, json, or text formats. Server logs use a clean structured layout with timestamps, levels, logger context, and optional JSON fields.

Organizations can now configure trusted email domains to streamline teammate discovery and invitations.

The organizations list now properly loads all entries with pagination support, ensuring large deployments can access every organization.

Personal account settings are now consistently accessible regardless of navigation context.

Sharing a service connector with Read or Edit access now correctly grants permission to use it for authentication, not just view its metadata

matching the behavior already available for secrets. The access list now accurately reflects who can actually use each connector.

You can now run individual pipeline steps on Databricks with the new Databricks step operator. This is useful when you want specific steps to execute in the Databricks runtime while the rest of your pipeline uses a different orchestrator. The Databricks orchestrator also supports optional tag settings to label jobs and cluster resources for cost tracking, ownership, and governance.

Dynamic pipelines can now be nested, allowing you to call one dynamic pipeline from within another. This enables more modular and reusable pipeline designs, letting you compose complex workflows from smaller dynamic building blocks.

The Run:AI step operator now supports advanced training workload settings, including multiple mount types (PVC, ConfigMap, Secret, NFS, S3, HostPath), workload templates, security context settings (UID/GID, non-root execution, seccomp, capabilities), port declarations and external URL exposure, and training workload parallelism and completions.