npx @buildinternet/releases get thundersRecently shipped live browser viewing in every run, run takeover, and Thunders AI effort levels.
Watch a run and step in when it breaks. Every run now streams its browser live. When a step fails or hits a breakpoint, take control, add the missing step, and resume; recording in the live browser turns your actions into steps. A redesigned test steps panel shows each step's state on a timeline.
Thunders AI gained effort levels and profile handling. Effort levels run low to high, trading speed for thoroughness in test generation, and it offers authenticated profiles when a sign-in fails. Plan-first creation remains: agree on steps and defaults before it writes the test, with mobile prompts for mobile and web+mobile projects. Guided recovery still suggests what to do next after a failed browse or agent step.
Squash TM now imports, creates, and syncs. Connect it to import tests, create tests, and sync results back. Runs can be stopped through the public API with a token, and display options apply across all lists.
Runs and failures keep explaining themselves. Failed runs show a suggestion per failed step with the reason; Failures Avoided counts each selector break self-healing repaired. Smaller changes: mobile device language and IP location settings, CI-uploaded builds, bulk test-set assignment, proxy routing, device rotation steps, zip download of several test assets, and custom PDF report titles with the project name on the cover.
Self-hosted v2026.10.15 carries the same feature set as the October 15 cloud release.
Thunders AI now executes tests and debugs them until they pass, and users can take control of the live browser at a failed step and record the fix. Authenticated profile pools give each parallel run its own login session, alongside API app spec support and API apps for Thunders AI.
Thunders AI can execute tests and debug them until they pass, with higher effort allowing more retries, and lets you take control of the live browser or mobile device at a failed step. Skills provide reusable project-shared instructions on how a team tests, authenticated profile pools give each parallel run its own login session, and runs blocked by a missing value now show a Not Runnable status.
Runs can now be watched in a live browser, with the ability to take control when a step fails or hits a breakpoint, add the missing step, and resume. Also adds Thunders AI effort levels, a redesigned test steps panel with a per-step timeline, Squash TM integration for importing tests and syncing results, and authenticated profiles offered when sign-in fails.
Self-hosted release matches the October 15, 2026 cloud feature set: a live browser in every run with run takeover and in-browser recording, Thunders AI effort levels, a redesigned test steps panel, Squash TM integration, authenticated profiles in Thunders AI, display options across lists, and stopping test runs via public API token.
Thunders MCP adds a get_authoring_guide tool for step types and variable syntax, plus search_test_cases for searching by text, label, and status. Also adds an admin Playwright project export endpoint at /api/admin/script-export and ends stuck runs as Interrupted.
Thunders AI now shows a plan to review before it writes a test, with mobile prompts for mobile and web+mobile projects, and guided recovery for failed browse or agent steps. Thunders MCP gains get_authoring_guide and search_test_cases tools plus controls for step-generation cost.
Self-host admins can download a diagnostics ZIP from /api/v2/selfhost/diagnostics covering version, configuration state, and instance health, while Thunders MCP agents can now read run evidence, step screenshots, and test case history. Also adds context-rich suggestions for failed test steps, bulk test asset download, custom PDF report titles, bot protection and page error warnings, and fixes for live run page sync and Notion bug tickets.
Failed test runs now show a suggestion with the failure reason for each failed step, including runs from CI, and Thunders MCP lets AI agents read run evidence, step screenshots, and test case history. Also added bulk zip download of test assets, a custom PDF report title, bot protection and page error warnings, and device rotation as a mobile test step.
Thunders AI now resolves variables, credentials, and addresses from a chosen environment and app, and web browse opens that environment with its real execution settings. Authenticated Profiles capture a logged-in browser session for reuse across tests, and self-healing selector repairs are now surfaced per run with a Failures Avoided score in Analytics.
Thunders AI now resolves variables, credentials, and addresses from a chosen environment and app, and Authenticated Profiles let a captured logged-in browser session be injected into fresh tests. Analytics adds a Failures Avoided score and trend for self-healed selectors, alongside bulk-assign of test cases to test sets, proxy support for runs, and fixes to Visual Compare on vertically shifted pages and browser driver reconnects.
Thunders AI now resolves variables, credentials, and addresses from a selected environment and app, and Authenticated Profiles capture a logged-in browser state with Capture profile and Inject profile steps for reuse across tests. Self-healing selector repairs are now visible as a Failures Avoided score in Analytics, plus bulk test-case assignment, proxy support per run, and a PDF report summary page.
Mobile testing is generally available: add a mobile app to a project, upload a build, and record or author steps with Thunders AI, with device pools controlling where runs execute and warm devices skipping cold-start. Admins can now grant access to specific projects, and PDF test reports open evidence at full size and quote the differences behind failing Visual Compare verdicts.
Admins can now grant a person access to only specific projects, including at invite time, and self-hosted instances can connect to Notion, Linear, ClickUp, and Atlassian. Visual Compare accuracy fixes cover overlay sizing, side-by-side on rejection, and full-page screenshots over inner-scroll content, and new optional per-AI-family resilience and fallback model environment variables were added.
Admins can now grant a person access to only specific projects, including at invite time, and evidence in PDF test reports opens at full size with the exact differences behind a failing Visual Compare verdict. Thunders AI also accepts XLSX attachments for writing tests, and Thunders MCP can read an environment and its variables rather than only create them.
A new read-only project role is managed from a per-project access table, and tools like Web Browse, Jira, Xray, and Linear can be connected and enabled directly from the chat, where Thunders AI clarifying questions now appear in a dedicated dock. Also adds mid-run variable updates with staleness indicators in the run view, a delete_step MCP tool, and Xray auto-sync of pass/fail results.
Clarifying questions now appear in a dedicated chat dock, and tools like Jira, Xray, and Linear can be connected and enabled directly from chat. Variables can update mid-run and apply globally, a read-only role is added for project access control, and an RGAA 4.1.2 accessibility audit grid is available for download.
Thunders AI clarifying questions now appear in a dedicated dock with suggested answers, and tools like Web Browse, Jira, Xray, or Linear can be connected and enabled directly from chat. Also new: a read-only role for viewing projects without changes, an RGAA 4.1.2 audit grid in ODS format, per-project execution cluster restrictions, a Thunders MCP delete_step tool, and a payment failed retry banner.
Test runs can now be paused and resumed mid-test with breakpoints, with variables updatable mid-run and applying globally, and Xray auto-sync pushes results automatically on run completion. The release also adds per-environment default execution settings, Excel data files for batch execute, region-based mobile device routing, and PDF export with an evidence toggle, alongside fixes for test execution reliability, MCP custom connectors, and auth token renewal.
Breakpoints let you pause and resume runs, and run settings can now be configured per environment with web and mobile tabs. Also fixed failed token renewal, the Chrome breach dialog, conditional steps, steps added during runs, file field references, and custom connectors in Thunders MCP.
Runs can now be paused at breakpoints, executed, fixed, and resumed, and web and mobile run settings are configurable per environment. Each environment also defines a read-only reserved [ENVIRONMENT_NAME] variable, and batch execution accepts .xlsx files alongside CSV.