Release notes 0.1.185

3 November 2026: The create card lists the steps.

Release notes · Updated 2026-11-03 · Describes release 0.1.185

0.1.185 — 3 November 2026

The create card lists the steps. A workspace without a lab workspace sees what creating one does, and an owner can pick a write-role kubeconfig of the workspace to move into the new lab workspace in the same step.
Lab connections explain themselves. The Lab cluster card lists every kubeconfig with its role and, for one that cannot be the lab cluster, the reason (read role, disabled, not a kubeconfig). A Lab incidents link opens the lab workspace's incidents.
Earlier runs stay visible. Runs started on the parent workspace before the lab workspace existed are listed, read-only, inside the lab.

Mark cloud connections as lab from the page. A lab workspace whose cloud lane is refused because its AWS connections are not marked as lab shows Mark this workspace's cloud connections as lab.
Graded against a fixed table, not the finding itself. Each Security Hub finding is now graded against the expected family, remediation stance and tier for its control; before, the expected values were copied from the finding, so classification always passed.
49 replay cases for the non-compliant stack. The controls the lab's CloudFormation stack is built to fail ship as gen-tmpl-* replay cases. Deploying and deleting the stack stays a manual step, described in the Test Lab guide.

RDS parameter-group changes keep the full ARN. A harvested ModifyDBParameterGroup (and other RDS calls) records the resource ARN from the CloudTrail event, so the change links to the scanner's findings and the AWS Config before-state.

Compare models in one click. A dataset's Run dialog lists the gateway models under Compare models; ticking two or more starts one pinned run per model (up to 8) over the same cases. The launch is refused while any ticked model has no card price, and the refusal names those models.
The model report keeps sandbox and live apart. A sandbox run and a live run of the same model are now two rows, not one replacing the other. Each row adds the mode, whether the gateway keeps the model in APAC, and the median seconds per case.
Per-model table in the Test Lab. The Progress card shows the newest run of each model on the newest catalogue: pass rate, cost per case and lab time per case (the whole round, not model latency).
Two pinned lab runs no longer swap models. A model pin applies to the whole workspace, so a pinned run is refused (409) while another lab run of the workspace is running, and no run starts while a pinned one is.

More Bedrock models price from the AWS Price List. Claude Haiku 5.5, Nova 2.0 Lite and OpenAI GPT-6 Astra now map to their AWS price-list rows.
Every Bedrock model on the gateway is priced before its first run. The price sync now covers every Bedrock alias the gateway lists, not only the models already booked or configured.