Skip to content

2.1.725

Latest
Latest

Choose a tag to compare

View all tags
@arthur-louisa arthur-louisa released this 23 Jul 18:17
2.1.725
f0863bc
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

🚀 Arthur Engine Release

July 23, 2026

This release focuses on streamlined dataset workflows, clearer visibility into built-in evaluators, and significant reliability improvements for prompt experiments and ml-engine scaling.

Datasets & Traces

Bulk Trace Management

  • Added the ability to bulk add traces to a dataset directly from the trace table by selecting multiple traces and clicking "Add to dataset" in the selection toolbar, backed by a new /api/v2/datasets/{dataset_id}/bulk-add-traces endpoint (#1968)

Users can now select and add up to 25 traces to a dataset in a single action, with a completion toast confirming how many traces were successfully added—eliminating the need to add traces one at a time.

Dataset Editing Safeguards

  • Added a warning alert that notifies users when navigating away from a dataset with unsaved column configurations, preventing silent loss of work on empty datasets (#1934)

This change closes a gap where configured columns on datasets without rows could be lost without notice.

Evaluations & Experiments

Built-in Evaluator Clarity

  • Built-in evaluators in the eval detail view now display a "Built-in" pill next to the title and an informational note explaining that Arthur maintains and locks their detection logic (#1971)

Users can now easily distinguish Arthur's native ML evaluators—such as PII, toxicity, and prompt injection detection—and understand why their Instructions sections cannot be edited.

Prompt Experiment Reliability

  • Fixed two independent out-of-memory issues in prompt experiments by flushing test cases per dataset row at create-time and streaming summaries at run-time (#1965)

Experiments with large mapped dataset columns, such as RAG context, can now be created and run to completion without hitting out-of-memory errors.

User Experience

Trace Annotations

  • Improved readability of annotation explanations in the trace annotations modal with a new expandable table row component (#1959)

Users can now expand annotation explanations for better readability, reducing text crowding in the initial modal view.

Infrastructure & Deployment

ml-engine Autoscaling

  • Added a HorizontalPodAutoscaler to the arthur-ml-engine Helm chart, bringing it to feature parity with arthur-genai-engine (#1977)

Enabled by default, the HPA scales on CPU (50%) and memory (80%) utilization with configurable replica bounds (2–10), allowing the platform to adapt to varying job queue depths without manual replica edits.

Documentation

  • Corrected critical gaps in the AWS EKS + EFS model-store documentation, including broken Terraform references, a config-copy race condition, access-point UID/GID requirements, and EFS mount target constraints (#1972)

Users setting up EFS-backed model stores on EKS can now follow corrected end-to-end steps without encountering race conditions, mount failures, or permission issues.

Security

Both medium-severity vulnerabilities could previously cause excessive runtime or memory consumption; users are now protected against these attack vectors.

Dependency & Maintenance Updates

  • Updated azure-storage-blob to v12.30.0, improving blob download performance, retry reliability, logging consistency, and fixing a potential memory leak (#1973)
  • Updated authlib to v1.7.2 for improved OAuth/JWT handling, including fixes for non-RSA key signing and more flexible algorithm support (#1970)
  • Updated axios to v1.18.1 with fixes for circular JSON serialisation, Node HTTP adapter proxy handling, and type definition accuracy (#1967)
  • Updated arthur-client to v1.4.2268 across the genai-engine and ml-engine modules (#1966)
  • Updated alembic database migration tool to v1.18.5 (#1969)
  • Updated the black code formatter to v26.5.1, adding Python 3.15 syntax support and formatting stability fixes (#1974)
  • Removed accidentally-committed blazebot session-memory files from the repository and added blazebot/ to .gitignore to prevent future commits (#1981)

These updates keep the Arthur Engine's dependencies current, improve internal tooling, and strengthen overall stability and performance.

Release notes generated by Louisa

Assets 2
Loading