v2026.07.27
TL;DR
Run across multiple replicas without pinning traffic to one instance, get step-quality suggestions before and after every run, execute Test Plans the same way as Test Sets, rate Thunders AI answers from the chat, and pick up a CORS fix for the self-hosted web console.
FEATURE
- Multi-replica scale-out: real-time updates now work across multiple replicas without pinning traffic to a single instance, so you can scale out horizontally. See the deployment guide's SignalR transport section for the Redis backplane configuration.
- Step-quality suggestions: get suggestions and ready-to-apply rewrites before and after a run, for more stable tests. Now generally available.
- Bulk suggestions via Thunders MCP: read step-quality suggestions across many test cases at once and apply the ones that come with a ready rewrite, directly through the Thunders MCP server.
- Test Plans can now be executed: Test Plans inherit the same structure as Test Sets: they can be executed, they get an owner, and they show their own test cases.
- Test Set owners: every test set now has an owner, shown in the list and its properties panel, same as test cases. The creator becomes the owner by default and you can change it.
- Thumbs up/down feedback on Thunders AI: rate a Thunders AI message right from the chat, in both the test case editor and the global assistant.
IMPROVEMENT
- Test-case-level accessibility report: the accessibility persona can produce one report for the whole test case instead of one per step, when reporting mode is set to comprehensive.
- Console logs: the Console Logs button is greyed out when a run has no logs, instead of downloading an empty file.
FIX
- Self-hosted web console CORS: the self-hosted web console's origin is now allowed through CORS, so the console can reach the API from its configured URL.
- Step-edit conflicts: editing a step now flags when a different user changed the test case first, instead of silently overwriting their work.
- Drag-and-drop reliability: clicks and drags now land on the correct target, including native HTML5 drag widgets.
- Duplicate step execution: a step no longer runs twice by replaying a stale result from a previous attempt.
- Steadier test runs: more reliable screenshot capture, natural language step recognition, Visual Compare detection, and faster failure when a browser session is lost.
Ready to ship faster with smarter testing?
