Lighter widget: the script-tag SDK's first download drops from about 243 KB to 119 KB gzipped. The screenshot and replay libraries load from next to testosaurus.js only when a screenshot is taken or replay starts; if they cannot load, the report still sends without them.
The bot check asks once more when its frame reloads or gives no answer in twenty seconds, instead of failing the report straight away.
Session replay records dropdowns (<select>) as blocked boxes instead of showing their first option.
A more honest widget: reports carry a per-report id so a resend is never counted twice, requests time out instead of hanging, consent works when the browser blocks storage, the launcher stays reachable in dialogs, and a missing apiKey now logs a warning. The quickstart gains a Content Security Policy section.
Safer reports: secrets in page addresses, console output and text are redacted before a report leaves the browser, typed text is masked in screenshots and replays, and data-testosaurus-mask and data-testosaurus-block mark anything else. Hidden characters in report text are shown as visible markers.
Privacy Policy updated, effective 2026-10-17: when you connect a coding agent, the approximate city and country of the request, the approval and the pickup are kept for at most 7 days to protect against approval phishing, deleted data can remain in encrypted backups for up to 30 days before the backups expire, and during sign-up we set a cookie for one hour so that the confirmation link signs in only the browser you signed up in. The Terms of Service are unchanged.
Docs: a new Connect your coding agent page covers connecting Claude Code and Codex over MCP, the fix prompt, key scopes and the safety model, and the quickstart gains a fourth step plus script-tag steps for Next.js and Vite. The landing page and the pricing cards now describe the agent loop and what each plan includes.
Agent loop over MCP: list_issues sorts by impact (or severity, reports or recent) and shows each issue's fix state, record_fix writes the result of a fix back to an issue, the fix_top_issue and triage_inbox prompts run the loop, and a Triage key scope reads everything and calls record_fix with no other write.
One search in the dashboard: Cmd/Ctrl+K or / opens a palette that finds issues, projects, pages, settings and actions.
Reports skipped for lack of credit are analyzed again right after a top-up, and on the next daily run once the workspace has credit.
A report you resolve or dismiss before it is grouped stays closed instead of joining an open issue, and bulk resolve works when grouping lands in between.
Members see whether AI analysis is on for a project; a partly failed bulk action says why.
Issues instead of piles of reports: reports about the same problem now share one issue with one analysis, a report count and merge and split controls. Every original report stays available, and an uncertain match is shown as a possible match instead of being merged.
Coding runs: give a coding role a model, a branch, the paths it may edit and your test commands, install the reviewed GitHub Actions workflow, and choose Propose fix on a distilled report. The agent works inside your GitHub Actions, every configured test must pass, and the result is a draft pull request you review, revise with feedback and merge yourself. Runs are paid from prepaid credit on every plan, commits carry [skip ci], and automatic runs are opt-in per category and capped.
Coding agents over MCP: 30 tools, each marked read-only, destructive or open-world, and a Connect-an-agent flow in Workspace settings that sets an agent up with one prompt. A key made without an intent is read-only. Visitor text comes back under untrusted_visitor_content, issue-changing tools no longer return the analysis text (read it with get_issue), coding-run tools list each test as its command, exit code and output size (get_coding_run adds the last 4,096 characters of each test's output), the configuration document carries each project's id and keeps your widget tokens unless you ask to prune them, and keys stay out of config files.
Verified email: new accounts confirm their address before they reach the dashboard, invitations survive the confirmation link, a password reset signs out other sessions, an email change is confirmed through the new address, and high-severity alerts go to verified addresses only.
Teams: teammates you invite can view and triage reports; only the workspace owner changes settings, connections, tokens and billing. Invitations come with a link you can copy, seats are counted, and a teammate can leave the workspace.
Billing you can follow: optional automatic top-ups with a monthly limit, monthly credit on the Pro invoice, and every payment listed with its fiscal invoice in Invoice history. Deleting an account now cancels billing first and asks you to confirm that unused purchased credit is forfeited. Referral links and discount invitations are paused for now.
Allowed domains take one hostname per entry; reports are checked against the page they came from, and the Installation screen says why a report was rejected.
Your data and its lifecycle: reports and uploads are purged after 90 days on Free and 365 days on Pro, uploads that no report claims after 48 hours, and deleting a report removes the tester's words from the shared analysis, which is then marked outdated. Beta access now runs to the end of the UTC month in which its 60 days end, with reminders at 14, 3 and 0 days.
Terms of Service and Privacy Policy updated, effective 2026-10-10: invite codes are optional, every metered resource and the automatic refill authorisation are covered, and the policy names Vercel AI Gateway, StartCo and ANAF e-Factura, the coding-run flow, retention, the browser storage keys and Cloudflare Web Analytics. The free-tier models left the model catalog, so the promise that your data is never used to train models is now absolute.
The documentation was rewritten to match the product: quickstart, troubleshooting with the exact console messages, API reference, dashboard guide, pricing and the MCP guide. The build now checks that every documentation link resolves.
Pick your AI: a new App settings → AI & routing tab opens the chat-model catalog of the Vercel AI Gateway — assign models by role (a default workhorse, a vision model for screenshot submissions, custom “expert” roles you name), let high-value reports route themselves to the right expert automatically, and see each model’s per-token price and typical cost per report before choosing.
AI billing meters input and output tokens at catalog base-token rates plus 5%, preserving fractional rates until the final charge. Plans include monthly AI credit (Free €1, Pro €3); optional paid overage draws from a prepaid balance. A daily spend cap and per-model usage breakdown help you track spending. Skipped reports are not automatically reprocessed after a top-up.
Free now includes 1 GB of screenshot and replay storage. Pro stays €29/month with 25 GB storage. Larger requirements use agreed Enterprise allowances and support terms.
Correction: the PagerDuty, Jira and Opsgenie integrations announced in v0.4.0 never shipped, and Pro does not include them. Issue routing works with GitHub and Linear.
Correction: the project export announced in v0.6.0 contains every submission with its distilled insight and a manifest of its screenshots and replays with their fetch paths, not the media files themselves.
Bring your team: invite teammates by email from Dashboard → Workspace settings → Team, and they accept through a signed link that checks their identity on arrival.
Bot protection got quieter: new projects get an invisible Turnstile widget, and existing projects can switch from the interactive checkbox to invisible in App settings → Installation → Advanced bot protection — the challenge now expands only while Cloudflare needs it.
Your data stays yours: one-click export per project downloads every submission with its distilled insight, screenshots, and replays.
The usage page gained an activation table per project — first submission, first distilled insight, and the time between them — so you can see time-to-value, not just volume.
Billing got easier to read: a reworked billing page, clearer usage meters, low-balance email warnings with a balance strip in the dashboard, and overage billing is now off unless you switch it on per resource. Top-ups start at €5.
Integration settings moved onto dedicated cards in Connections and App settings → Issue routing with live connection status.
Fixes: usage alerts now reach every workspace owner instead of colliding on shared keys, and the Linear token exchange and refresh use the form encoding Linear's API expects.
Capture consent got friendlier: the widget now opens with a compact wizard step inside the panel where testers accept the screen recording, fall back to a screenshot only, or skip it entirely — then land straight on the feedback form.
You choose how consent is presented, per project, in Dashboard → App settings → Widget: the new wizard, inline opt-in checkboxes in the form, or the original separate consent cards. Included on every plan, including Free.
Make the widget yours: pick an accent color, switch between the Testo launcher and a minimal chat bubble, go dark or light, and write your own panel title, input placeholder, welcome message, and button label — with up to 6 of your own tags. Edit everything in Dashboard → App settings → Widget and watch the live preview update as you type.
Position, dark or light surface, and launcher choice are included on every plan. Brand colors, custom wording, and custom tags are part of Pro's unlimited customization.
Usage-based billing: every plan includes monthly allowances for submissions, storage, distillations, and team seats, with prepaid overage instead of surprise bills. Track it all on the Usage screen.
PagerDuty, Jira, and Opsgenie integrations are now included with Pro and Enterprise.
Free plans carry a "Powered by Testosaurus" badge; Pro and Enterprise can remove it in project settings.
Put agents to work on your feedback: a remote MCP server, management keys (read-only or writable), a one-time Connect-an-agent bootstrap flow, and a config API.
Manage screenshots and replays from the dashboard, the REST API, or MCP: browse, delete, and bulk-clear by project.
Insights gained category, severity, and status breakdowns, plus per-project distillation settings.
The marketing site's Testo now points at the widget.