ZenML Release Notes

Follow

131 release notes curated from 2 sources by the Releasebot Team. Last updated: Nov 7, 2026

Get this feed:
  • Nov 6, 2026
    • Date parsed from source:
      Nov 6, 2026
    • First seen by Releasebot:
      Nov 7, 2026
    ZenML logo

    ZenML

    Bots are live, with a refreshed UI

    ZenML Pro adds always-on AI engineer Bots for investigating runs, testing fixes, and opening pull requests, alongside a refreshed UI.

    ZenML Pro now has Bots, an always-on team of AI engineers. Ask one to investigate a failed run or an agent's behavior, test a fix, and open a pull request, from the UI, Slack, or GitHub. You can also put recurring reviews on a schedule.

    The UI has a new look too, with the same runs, models, and artifacts. Prefer the old one? It's still at classic.cloud.zenml.io.

    Original source
  • Nov 4, 2026
    • Date parsed from source:
      Nov 4, 2026
    • First seen by Releasebot:
      Nov 5, 2026
    ZenML logo

    ZenML

    Rename projects from the CLI and SDK

    ZenML now lets users rename projects from the CLI while preserving their UUID and resources, and safely update or clear SDK descriptions.

    You can now rename a project from the CLI with zenml project update <name-or-uuid>, using --name, --display-name, and --description. The technical name and the display name can be changed separately, and the project keeps its UUID, so its runs, pipelines, and other resources stay attached. Updating a project from the SDK now also leaves fields you don't pass unchanged, and lets you explicitly clear the description.

    Original source
  • All of your release notes in one feed

    Join Releasebot and get updates from ZenML and hundreds of other software products.

    Create account
  • Nov 4, 2026
    • Date parsed from source:
      Nov 4, 2026
    • First seen by Releasebot:
      Nov 5, 2026
    ZenML logo

    ZenML

    Helm chart pods meet Pod Security `restricted` by default

    ZenML updates its Helm chart so all created pods meet Kubernetes’ restricted security profile, with guidance for safe upgrades in locked-down namespaces.

    Every pod the ZenML Helm chart creates now satisfies the Kubernetes Pod Security restricted profile out of the box: the server, the database migration job, workers, CA certificate init containers, and the helm test pod. You no longer need extra overrides to deploy into locked-down namespaces.

    When you upgrade, use --reset-then-reuse-values (Helm 3.14+) instead of --reuse-values. With --reuse-values, Helm keeps the defaults from the chart version you first installed, and in a restricted namespace the upgrade fails because the database migration pod is rejected.

    Original source
  • Nov 4, 2026
    • Date parsed from source:
      Nov 4, 2026
    • First seen by Releasebot:
      Nov 5, 2026
    ZenML logo

    ZenML

    More reliable installs, GCS runs, and replays

    ZenML fixes integration dependency conflicts and GCS upload stalls, improves artifact handling and replayed-run parameters, and lets CLI labels include equals signs. MLflow integration also installs on Python 3.13 and slim images without a C++ compiler.

    Installing a ZenML integration no longer upgrades shared packages into versions that break ZenML's own imports, GCP runner images no longer pull in a conflicting OpenTelemetry version, and the MLflow integration now installs on Python 3.13 and slim images without needing a C++ compiler.

    Steps using a GCS artifact store no longer hang during final uploads or cleanup, and two new optional settings, request_timeout and read_cache_type, help if you still see stalls.

    ZenML also keeps step outputs when a step fails after its artifacts were recorded, replayed runs now pass along parameters the original run took from defaults, and CLI --label values can now contain =, such as URLs with query strings.

    Original source
  • Nov 4, 2026
    • Date parsed from source:
      Nov 4, 2026
    • First seen by Releasebot:
      Nov 4, 2026
    ZenML logo

    ZenML

    0.98.1

    ZenML releases updates that make project renaming easier and improve reliability across integrations, pipeline replays, GCS storage, and locked-down Kubernetes deployments. The release also fixes CLI label handling and hosted dashboard access, helping teams run and manage pipelines more smoothly.

    Added

    Rename projects from the CLI and SDK: You can now rename projects with zenml project update <name-or-uuid> using --name, --display-name, and --description (PR #5367). Technical names and display names can be changed independently while preserving the project UUID and associated resources, and SDK updates now correctly preserve omitted fields or explicitly clear descriptive fields.

    Helm chart pods now satisfy restricted Pod Security defaults: The default ZenML Helm chart pod security settings now work in namespaces enforcing Kubernetes Pod Security restricted (PR #5397). This also applies the chart security values consistently to CA certificate init containers and Helm test pods, reducing the extra overrides needed for locked-down clusters.

    Fixed

    More reliable integration installs: Installing ZenML integrations is now less likely to break ZenML’s already-installed dependency set (PR #5399). This prevents cases where adding an integration upgrades shared packages into incompatible versions and causes import failures before pipelines can run.

    GCP runner OpenTelemetry compatibility: GCP runner images now avoid installing an incompatible OpenTelemetry API version alongside ZenML’s pinned SDK (PR #5366). This fixes import failures that could happen before any pipeline steps started.

    MLflow integration installs on Python 3.13 and slim images: The MLflow integration no longer depends on the unused python-rapidjson package (PR #5360). This avoids source compilation failures in environments without build tools such as g++, especially on Python 3.13 or slim container images.

    GCS-backed runs no longer hang during final uploads or cleanup: ZenML aligned GCS-related dependencies and added optional storage controls to avoid pipeline steps getting stuck after their work and outputs have already completed (PR #5373). This addresses hangs from indefinite GCS responses and affected gcsfs cleanup behavior, helping orchestrators observe completed containers instead of waiting forever.

    Step outputs are preserved after execution or finalization failures: ZenML no longer deletes step output directories from the launcher exception handler when a step fails after artifacts or server-side completion have already been recorded (PR #5362). Persisted metadata will no longer point to missing files in these partial-failure scenarios, while the original error is still propagated.

    Replay runs include defaulted pipeline parameters: Replaying a run now forwards pipeline parameters that were satisfied by defaults in the original run (PR #5361). This keeps replayed runs consistent with the source run even when users did not explicitly pass every parameter.

    CLI labels can contain equals signs: CLI label parsing now splits --label values only on the first =, so values such as URLs and query strings are accepted (PR #5383). This fixes label handling across service connector and stack component commands where values like https://example.com/?a=b were previously rejected or misinterpreted.

    Hosted Classic dashboard CORS origins are allowed automatically: Pro workspace servers now automatically include the hosted Classic dashboard origin in addition to their configured dashboard origin (PR #5396). Production and staging Classic dashboard URLs are matched exactly over HTTPS, while custom enterprise dashboards and existing custom CORS origins keep their current behavior.

    What's Changed

    • Add 0.98.0 to the migration tests by @github-actions[bot] in #5358
    • Add version 0.97.0 to legacy docs by @github-actions[bot] in #5356
    • Bump the minor-and-patch group across 1 directory with 4 updates by @dependabot[bot] in #5363
    • Removing rapidjson dependency from the mlflow integration by @bcdurak in #5360
    • Preserve step outputs when execution or finalization fails by @bcdurak in #5362
    • Prevent OpenTelemetry conflicts in GCP runner images by @bcdurak in #5366
    • Bump the minor-and-patch group across 1 directory with 5 updates by @dependabot[bot] in #5372
    • Skip unsupported Jev answer types when logging metadata by @bcdurak in #5375
    • Support project renaming in the CLI and SDK by @bcdurak in #5367
    • Bump the minor-and-patch group across 1 directory with 5 updates by @dependabot[bot] in #5387
    • Bump zizmor from 1.30.1 to 1.31.1 in /.github/zizmor by @dependabot[bot] in #5386
    • Fix CI failures caused by incompatible dependency releases by @bcdurak in #5385
    • Reuse one Modal App per component instead of creating one per run by @strickvl in #5311
    • Align GCS dependencies and add optional storage controls by @bcdurak in #5373
    • Allow = in CLI label values by @RaphaelFakhri in #5383
    • Automatically allow hosted Classic dashboard origins by @AlexejPenner in #5396
    • Forward defaulted pipeline parameters when replaying a run by @schustmi in #5361
    • Add Bots docs to ZenML Pro by @schustmi in #5371
    • Keep integration installs from breaking ZenML's dependencies by @bcdurak in #5399
    • Fix Windows, macOS and security-scan CI failures by @bcdurak in #5400
    • Prepare release 0.98.1 by @github-actions[bot] in #5398
    • Add version 0.98.0 to legacy docs (#5405) by @bcdurak in #5407
    • Fix heartbeat tests breaking the unit tests on pytest 9.2 (#5408) by @bcdurak in #5409

    New Contributors

    • @RaphaelFakhri made their first contribution in #5383

    Full Changelog: 0.98.0...0.98.1

    Original source
  • Similar to ZenML with recent updates:

  • Oct 27, 2026
    • Date parsed from source:
      Oct 27, 2026
    • First seen by Releasebot:
      Oct 30, 2026
    ZenML logo

    ZenML

    Sync identity-provider claims into ZenML users (self-hosted Pro)

    ZenML now maps SSO identity claims to Linux UIDs and GIDs, aligning Kubernetes pipeline runs with organizational file permissions on shared storage.

    Self-hosted ZenML Pro administrators can now copy selected claims from their enterprise identity provider onto ZenML user accounts, using new Helm values (zenml.auth.sso.uidClaim, gidClaim and extraClaims). Map a Linux UID and GID claim, and pipelines launched with the Kubernetes orchestrator or step operator on ZenML 0.98.0 or later run as that same Linux user and group, so file permissions on shared storage line up with your organization's accounts. Users pick up their claims on their next SSO login.

    Original source
  • Oct 14, 2026
    • Date parsed from source:
      Oct 14, 2026
    • First seen by Releasebot:
      Oct 23, 2026
    ZenML logo

    ZenML

    Match any webhook event, and safer snapshot deletion

    ZenML adds flexible webhook matching, future-ready trigger snapshots, and warnings before deleting snapshots that triggers still depend on.

    Webhook triggers can now match a provider's raw event type and any field in the authenticated JSON payload, so you can react to events that ZenML's typed event catalog doesn't cover yet, such as only newly opened GitHub pull requests that target a specific branch. Snapshots updated with this release also resolve trigger types added in future ZenML versions, so you no longer need to refresh them just to use a new trigger type. ZenML now also warns you before you delete a snapshot that a trigger still depends on.

    Original source
  • Oct 14, 2026
    • Date parsed from source:
      Oct 14, 2026
    • First seen by Releasebot:
      Oct 23, 2026
    ZenML logo

    ZenML

    Store Jev failure triage and run summaries as run metadata

    ZenML adds TypeSafe Jev hooks for failure classification and structured run summaries, stored as searchable run metadata.

    A new typesafe integration adds two hooks backed by TypeSafe's Jev model.

    Attach jev_failure_triage_hook as an on_failure hook to classify each failure (for example infrastructure, data, code, dependency, or credentials) with a probability, and jev_run_summary_hook to record structured run summaries.

    The answers are stored as ZenML run metadata, so you get searchable failure triage on your runs without writing custom metadata code.

    Original source
  • Oct 14, 2026
    • Date parsed from source:
      Oct 14, 2026
    • First seen by Releasebot:
      Oct 23, 2026
    ZenML logo

    ZenML

    Kubernetes workloads carry your organization's identity

    ZenML adds workspace identity annotations to Kubernetes workloads, supports synchronized Linux user and group IDs for enterprise deployments, and lets sandbox sessions be tied to their owners for automatic cleanup.

    Kubernetes Jobs, pod templates, scheduled and dynamic executions, and Run:ai training workloads now carry a zenml.io/workspace-user-id annotation with the workspace account that launched them, which makes audit and platform tooling easier to wire up. ZenML Pro deployments with an enterprise identity provider can also pass synchronized linux_uid and linux_gid claims into Kubernetes orchestrator and step operator workloads, so processes run as the matching organization Linux account on shared filesystems. Kubernetes-managed sandbox sessions can now be given owners, so they are cleaned up with the step they belong to instead of being left orphaned.

    Original source
  • Oct 14, 2026
    • Date parsed from source:
      Oct 14, 2026
    • First seen by Releasebot:
      Oct 23, 2026
    ZenML logo

    ZenML

    Transformers v5 and wider LangChain support

    ZenML expands Hugging Face compatibility to Transformers v5 and broadens the supported LangChain 0.x version range.

    The Hugging Face integration now supports Transformers v5 while staying compatible with v4, and the Accelerate integration uses a CPU configuration that works across both versions.

    The langchain integration now accepts langchain>=0.3.5,<1.0 instead of pinning 0.3.0. LangChain 1.x is not supported yet, so if you pin an exact LangChain version, update your constraints to fall inside the new range.

    Original source
  • Oct 14, 2026
    • Date parsed from source:
      Oct 14, 2026
    • First seen by Releasebot:
      Oct 23, 2026
    ZenML logo

    ZenML

    Pipeline reliability fixes

    ZenML fixes repeated cached DataFrame and NumPy loading in dynamic pipelines and concurrency failures in remote-orchestrator deployments.

    Loading the same cached pandas DataFrame or NumPy array more than once inside a dynamic pipeline no longer fails with "The truth value of a DataFrame is ambiguous". Concurrent /invoke calls against deployments that use remote orchestrators such as Kubernetes no longer race and fail with RuntimeError: Run context not initialized.

    Original source
  • Oct 14, 2026
    • Date parsed from source:
      Oct 14, 2026
    • First seen by Releasebot:
      Oct 15, 2026
    ZenML logo

    ZenML

    0.98.0

    ZenML releases updates that expand webhook triggers, improve snapshot safety, and strengthen Kubernetes workload identity and ownership. The release also adds Transformers v5 and Jev integration support, broadens LangChain compatibility, and fixes artifact caching and concurrent pipeline invocation issues.

    Breaking Changes

    The langchain integration now supports langchain>=0.3.5,<1.0 instead of the previously pinned 0.3.0. If you rely on a fixed LangChain version, update your environment constraints accordingly, and note that LangChain 1.x is still not supported by ZenML and may require you to stay on a compatible pre-1.0 release. PR #5341

    Triggers and automation

    • Dynamic webhook event filters: Webhook triggers can now match a provider’s raw event type and fields directly from the authenticated JSON body. This gives you an escape hatch for events or payload fields that ZenML’s typed event catalog does not expose yet, such as matching newly opened GitHub pull requests targeting a specific branch. PR #5247
    • Forward-compatible run.trigger resolution: Snapshots updated with this release can resolve trigger types that are introduced in future ZenML versions. This reduces the need to refresh snapshots just to use newly added trigger types such as platform or webhook triggers. PR #5281
    • Safer snapshot deletion for triggered pipelines: ZenML now checks whether snapshots are attached to triggers before deletion and warns users about unintended removals. This helps prevent accidentally breaking automations that still depend on a snapshot. PR #5284

    Kubernetes, Run:ai, and workload ownership

    • Synchronized OIDC Linux identities for Kubernetes workloads: ZenML Pro deployments using an enterprise identity provider can now carry synchronized linux_uid and linux_gid claims into workloads created by the Kubernetes orchestrator and Kubernetes step operator. This keeps the process UID and primary GID aligned with organization-managed Linux accounts, which is especially useful for shared filesystems and cluster security policies. PR #5301
    • Workspace account annotations on Kubernetes and Run:ai workloads: Kubernetes Jobs, Pod templates, scheduled and dynamic executions, and Run:ai training workloads now include a resolvable zenml.io/workspace-user-id annotation. This gives platform and audit tooling the actual workspace account UUID while preserving custom annotations and leaving ZENML_USER_ID unchanged. PR #5326
    • Owners for managed Kubernetes sandboxes: Kubernetes-managed Sandbox sessions can now be assigned owners. This makes sandbox sessions that are closely related to a step easier to track and helps avoid leaving orphaned sessions behind. PR #5336
    • Workspace service account activation controls: Activation and deactivation are restored for deprecated workspace service accounts and legacy API keys, including keys retained after adoption. This supports migration cleanup and credential rollback while preserving the existing deprecation restrictions and deletion checks. PR #5340
    • Transformers v5 support in the Hugging Face integration: The Hugging Face integration now supports Transformers v5 while retaining compatibility with v4. ZenML only registers the TensorFlow model materializer when running on Transformers v4, and the Accelerate integration now uses cross-version CPU configuration so users can upgrade their Hugging Face stack more smoothly. PR #5321
    • TypeSafe Jev hooks integration: A new typesafe integration adds jev_failure_triage_hook and jev_run_summary_hook for storing Jev’s typed, probability-scored answers as ZenML run metadata. Users can attach structured failure triage and run summaries to their runs without building custom metadata plumbing. PR #5312

    Fixed

    • Artifact cache loading for pandas and NumPy artifacts: ArtifactVersionResponse.load() no longer evaluates cached artifact data in a boolean context, which could raise errors like “The truth value of a DataFrame is ambiguous” when loading cached DataFrames or arrays. Loading the same artifact multiple times inside dynamic pipelines now works correctly for DataFrames, NumPy arrays, and other falsy values. PR #5272
    • Concurrent deployed pipeline invocations with remote orchestrators: ZenML no longer lets per-invocation step runner cleanup clear the deployment service’s process-wide run context. Concurrent /invoke calls against deployments that use remote orchestrators such as Kubernetes should no longer race and fail with RuntimeError: Run context not initialized. PR #5159

    What's Changed

    • Add 0.97.0 to the migration tests by @github-actions[bot] in #5304
    • Add version 0.96.4 to legacy docs by @github-actions[bot] in #5302
    • Recompute next occurrence when a one-off schedule is retimed by @joaquinhuigomez in #5293
    • Run Kubernetes workloads with synchronized OIDC identities by @stefannica in #5301
    • Forward compatible run.trigger resolution by @Json-Andriopoulos in #5281
    • Retry RDS IAM database connections while IAM changes propagate by @safoinme in #5294
    • Bump the minor-and-patch group across 1 directory with 5 updates by @dependabot[bot] in #5318
    • Fix artifact cache lookup in ArtifactVersionResponse.load by @kounelisagis in #5272
    • Fix context manager docstring linting by @AlexejPenner in #5327
    • Fix Claude Code pre-commit formatting hook that never ran by @strickvl in #5339
    • Fix run context teardown during concurrent deployment invocations by @kounelisagis in #5159
    • Allow to set owners for sandboxes by @avishniakov in #5336
    • Support for dynamic target events by @Json-Andriopoulos in #5247
    • Allowing workspace-level service account activation/deactivation by @bcdurak in #5340
    • Bump PyJWT to 2.15.x to fix medium-severity advisories by @strickvl in #5328
    • Fix broken AWS Strands and Semantic Kernel links in agent framework examples README by @pratikgx in #5342
    • Bump the minor-and-patch group across 1 directory with 5 updates by @dependabot[bot] in #5330
    • Add workspace account annotations to Kubernetes and Run:ai by @safoinme in #5326
    • Relax langchain integration pin to >=0.3.5,<1.0 by @strickvl in #5341
    • Support Transformers v5 in Hugging Face integration by @AlexejPenner in #5321
    • Trigger check before snapshot deletion by @Json-Andriopoulos in #5284
    • Add TypeSafe Jev hooks integration (jev_failure_triage_hook, jev_run_summary_hook) by @strickvl in #5312
    • Prepare release 0.98.0 by @github-actions[bot] in #5353

    New Contributors

    • @pratikgx made their first contribution in #5342

    Full Changelog: 0.97.0...0.98.0

    Original source
  • Sep 23, 2026
    • Date parsed from source:
      Sep 23, 2026
    • First seen by Releasebot:
      Sep 26, 2026
    ZenML logo

    ZenML

    Resource pools now run on the ZenML Pro Resource Manager

    ZenML rebuilds resource pools on the new ZenML Pro Resource Manager, adding allocation-based scheduling for dynamic steps and Kubernetes support for CPU, memory, and GPU pod requests and limits. It also keeps running steps alive with heartbeats and removes the old resource pool system.

    Resource pools have been rebuilt on the new ZenML Pro Resource Manager, which now decides scheduling, allocation, and eviction. Eligible dynamic steps request resources through it and wait for an allocation before they start, and Kubernetes is the first backend to apply the allocated CPU, memory, and GPU capacity as pod requests and limits. Running steps keep their allocation alive through step heartbeats and are cancelled gracefully if their resources are reclaimed.

    Breaking: the previous resource pool implementation has been removed together with its data — existing pools, policies, and resource requests are not migrated and need to be set up again.

    Original source
  • Sep 23, 2026
    • Date parsed from source:
      Sep 23, 2026
    • First seen by Releasebot:
      Sep 26, 2026
    ZenML logo

    ZenML

    Prune unused pipeline snapshots

    ZenML adds snapshot pruning and atomic creation to clean up unused pipeline snapshots and avoid partial leftovers.

    Unnamed pipeline snapshots left behind by deleted runs, deployments, and schedules can now be cleaned up. Run zenml pipeline snapshot prune --older-than-days 90 (also available as zenml prune snapshots) to delete snapshots that nothing references anymore, or add --dry-run to see how many would go first. Replacing a named snapshot with replace=True now deletes the old one when it is unused, and snapshot creation is atomic, so a failed attempt no longer leaves a partial snapshot behind.

    Original source
  • Sep 23, 2026
    • Date parsed from source:
      Sep 23, 2026
    • First seen by Releasebot:
      Sep 26, 2026
    ZenML logo

    ZenML

    Paginated, filterable logs for external log stores

    ZenML adds paginated external log reading with filters and a revamped Datadog log store to reduce rate-limit pressure.

    Reading logs from external log stores no longer means one huge query or hitting the backend's rate limits. A new GET /api/v1/logs/{logs_id}/entries endpoint returns logs in pages with cursors and supports search, level, since, and until filters, and the Datadog log store has been rewritten on top of it. If you maintain a custom log store, update it to the new BaseLogStore.fetch() signature and import LogEntry from zenml.models. For Datadog, the existing run and step log endpoints now return the newest 1,000 entries instead of up to 50,000.

    Original source
Releasebot

Curated by the Releasebot team

Releasebot is an aggregator of official release notes from hundreds of software vendors and thousands of sources.

Our editorial process involves the manual review and audit of release notes procured with the help of automated systems.