v2026.09.17
TL;DR
Admins can now restrict a person’s access to specific projects, a PDF test report links to full-size evidence and quotes the exact Visual Compare differences, and a self-hosted instance can now connect to Notion, Linear, ClickUp and Atlassian.
FEATURE
- Per-project access control: Admins can now give a person access to only specific projects, including at invite time.
- PDF report evidence: Evidence in a PDF test report now opens at full size in the browser. The report also quotes the exact differences behind a failing Visual Compare verdict.
- Write tests from your Excel files: Thunders AI can now accept an XLSX file attachment, not only CSV.
- Thunders MCP reads environments: Thunders MCP can now read an environment and its variables, not only create them.
- Connect Notion, Linear, ClickUp and Atlassian: A self-hosted instance can now connect Notion, Linear and ClickUp.
IMPROVEMENT
- Custom MCP Connector: A custom MCP connector can now be picked as a bug report target, the same as a built-in one.
- RGAA audit grid descriptions: The "Modifications à apporter" column now pairs each selector with a plain-text description of the element.
- List pagination in the URL: A list's page number now stays in the URL, so a refresh or a shared link returns to the same page.
- Search data files from a test: You can now search for a data file when you launch a test run from a file.
- Test steps panel controls: The substep controls now sit in a bottom bar and stay usable while a run is in progress.
- Thunders MCP stays connected: A Thunders MCP connection now stays signed in longer, so it needs to reconnect less often.
- Recording shows in test case activity: A recording batch now shows up in a test case's activity feed.
FIX
- Visual Compare accuracy: Fixed the overlay sizing so it never covers the element, added a clear message when a Visual Compare reference names no test asset, kept the baseline and capture side by side when a comparison is rejected, and Full Page screenshots and Visual Compare now expand the viewport over inner-scroll content.
- File Compare and CSV assets: Fixed File Compare reporting a pass when it found nothing to compare, and blank rows or trailing blank columns breaking a CSV test asset.
- Test case steps: Fixed a test case step being allowed to reuse the test case it lives in.
- Thunders AI question dock: Fixed enabling one suggested tool disabling another, a dropped reply when the underlying model's stream ended early, and an occasional plan for a step the runner cannot execute.
- UI polish: Fixed inconsistent list empty states, untranslated navigation button labels, the upload menu being unreachable by keyboard, user column widths resetting on navigation, and renamed the misleading "Total" analytics KPI labels.
- Recorder: Fixed the address bar not showing the recorded page's URL, and a merge-group ownership race.
- Self-hosted storage and providers: A missing self-hosted S3 blob is now reported as not found instead of crashing the request, and a self-hosted instance now survives an upstream AI provider capacity error instead of failing the request.
Ready to ship faster with smarter testing?
