---
title: Oct 2, 2026
slug: therafi-release-notes/oct-2-2026
createdAt: 2026-11-03T12:34:44.690Z
---

Changes released on Oct 2, 2026.

## New

- **Customizable Playwright analytics dashboard.** The Playwright dashboard can now be customized with widgets, and your widget preferences are kept separate from other dashboards.
- **PDF export for run reports.** You can now export a run report as a PDF directly from the report page.
- **Playwright API keys and reference.** Playwright API keys can be managed in settings, and the API reference now points to the real production hosts. The 100-scenario limit note has been removed from the create-run docs.
- **Minimum OS compatibility for mobile runs.** RafiRun now checks a test's minimum OS requirement when you pick a device. If the minimum OS cannot be verified, you can still select the device and run.
- **Interactive Recorder with a live sandbox.** The new Recorder panel streams a sandbox browser that you drive yourself, with checks you can point at on the page.

## Improvements

- **Clearer Recorder progress.** The Recorder shows startup milestones over the browser pane, an "asking the LLM" line while the model is working, and a "finishing" state once every step has a result. A failed attempt now appears as still trying, and reopened steps show the kind of action that actually ran.
- **Plain-language model failure messages.** When a model cannot be used, the wizard now says whether it is unreachable, rejected, unpaid, missing, throttled or erroring, in plain words, instead of asking you to review your prompt. The message also points to Modules > Playwright in the admin panel, or your administrator.
- **More reliable model routing for authoring.** For OpenRouter models, the provider routing set by the author is now stored and applied, and the first authoring job after a model assignment is routed correctly. Saving the Playwright module in the admin panel no longer overwrites the computed routing.
- **Better scenario names, descriptions and start URLs.** A scenario name is no longer pre-filled with a truncated prompt; the run's own name is used when it proposes one. A suggested description can replace an earlier suggestion but never text you typed. Local Recorder start URLs without a scheme open over http, and unreachable start pages are reported.
- **Security and access hardening.** Access is now scoped more tightly: company owners are limited to their own company, data and live events are scoped per company, and new API keys cannot exceed the scopes of the key that created them. Logging out revokes your session, 2FA is locked after repeated failed attempts, and share tokens are masked in logs. A temporarily unavailable session store is now reported as a service issue rather than a sign-in failure.
- **ECS capacity management for administrators.** Administrators get simplified ECS scaling settings and can start and stop temporary nodes from the interface.
- **AI Playground retired.** The AI Playground feature has been removed from RafiRun.

## Fixes

- **Test nodes keep their connection when reconnecting.** A node that reconnects within the disconnect grace period is now kept, and a node re-registering no longer closes its own connection or gets overridden by an old, closing one.
- **More dependable run and test reports.** Run status and test report updates are now applied safely, so concurrent updates no longer overwrite each other or move a finished scenario back to an earlier state. Run report access and deletion are also limited to your own company.
- **Interface polish.** Fixed dashboard KPI cards being clipped on narrow grids, inverted delta colors and the chart focus ring. The wizard's Live Stream keeps its height when the prompt card opens, the bulk invite dialog scrolls, and the selected folder stays selected after you create a scenario folder.