Background browser recovery
Let a background browser habit recover a changed route, then review its proposed steps before saving them.
last updated
Background browser recovery belongs to the experimental browser feature, which is disabled by default. It helps an eligible browser habit find a changed route or target. It is on by default for browser habits. It operates inside the same temporary browser and the websites reviewed with the habit.
Change the preference
In a browser habit's details, Allow AI recovery for this habit starts checked. Uncheck it to disable recovery; your choice survives restarting and syncing. Existing browser habits without a stored preference also start with recovery on after this update. Earlier versions did not distinguish an opt-out from an unset preference, so uncheck it again if you previously turned it off.
For the built-in example, open Debug → Chrome background automation and choose Test AI recovery, then review the normal run approval.
How recovery works
If a target cannot be found before an action is sent, Plainwork offers the model a small list of currently available controls. The model chooses a control, and Plainwork checks it before executing it under your approval for the whole run. It shows each recovery action in the live run panel without asking you again.
Recovery can take several steps, but it must satisfy the original goal. It stays in the same temporary browser and cannot open foreground Chrome or switch to Guide via Screenshots. Unknown action outcomes, account or permission failures, and unsupported operations stop the run.
Each run permits one recovery attempt, limited to five model turns, four actions and sixty seconds. Cancelling stops recovery and cleans up the browser.
What the model receives
Recovery sends the original goal and a bounded list of visible control identifiers and labels through Plainwork's existing cloud model service. It does not collect screenshots, full page text, raw DOM, field or hidden values, cookie records, storage dumps, or page URLs for recovery. Labels containing known session cookies or recorded field values are rejected, including common encoded forms. A website's visible labels can still contain sensitive text that the filter cannot recognize. Raw prompts and replies are not saved in the habit or run history.
Enabling recovery does not grant ongoing run approval or permission to import a Chrome session.
Review an improvement
After successful recovery, Review browser improvement shows the steps before and after. Accept improvement saves the validated steps; future successful runs replay them without a model. Reject improvement keeps the original habit. Use Replay saved habit to test the accepted fixture steps in a fresh browser.
If the habit changed since the proposal was created, acceptance stops rather than overwriting those edits. An unavailable account, protected-state failure, or failed save also prevents acceptance. Changing executable steps does not inherit permission to run automatically. If the server saved a change but the local save could not finish, Plainwork tells you to check sync before retrying.
The local qualification harness uses scripted choices. Its success does not establish live-model reliability or compatibility with your website.