How Dr. Rosie keeps the software running: automatic repair, security and reported problems

What watches and repairs Dr. Rosie’s own software around the clock, what is automatic and what is not, how security holes are watched, and what happens to a problem you report.

The short answer

Dr. Rosie is built and run by a small company. Part of the engineering watch is done by AI helpers that work around the clock: they check the software every minute, put the last working version back when a key page fails, and write the fix. People set the rules the helpers work inside, and are emailed what they did.

This article says exactly what is automatic and what is not. It is as of October 29, 2026.

When the software breaks: automatic repair

Every minute Dr. Rosie checks its own key pages: the health check, the home page, the Ask Dr. Rosie chat, a Learn article and the Learn hub, Community, the For Vets page and the Pricing page. When a check fails, the repair helper starts by itself. Nobody has to press anything.

  1. It puts the last working version back live, so clinics are not left on a broken one.
  2. An AI works out what broke and writes a fix.
  3. Other AI reviewers check the fix, and Dr. Rosie’s automatic tests must pass.
  4. If both pass, the fix goes live and a report is emailed to the company’s CEO. If they do not, it stops, leaves the working version live and emails that it stopped.

It was tested by deliberately breaking the live site three times, October 18 to 20, 2026. In the first test the working version was back live about three minutes after the failed check, and the fix was live within the hour. Those are test results, not a promise of how long any one problem takes.

Its limits: one automatic repair a day, a spending cap on each repair, and some kinds of change are off limits to it and wait for a person. It handles the site going down or a key page failing after a change. It is not a general bug-fixer.

Security

  • Every hour, Dr. Rosie compares the software it is built from against the US government’s list of security holes that attackers are actively using in real attacks. A match appears on the company’s monitor and is emailed to the CEO.
  • New versions of the parts Dr. Rosie is built from are not adopted until they have been public for seven days, so a bad release published today is not pulled in today. The exception is a fix for a known security hole.
  • An AI patcher fixes security holes by itself when the update stays within the same major version. It is switched on, but it is new and still being proven: its first live attempts did not finish and are being fixed. Until it works end to end, a person makes sure each hole is fixed.

Nothing here replaces a formal security audit. It is a way of finding known holes fast.

A problem you report

In Rosie, Support › Send Feedback takes bug reports and ideas. An AI reads each report, looks for similar earlier reports and for alerts that match, and writes a note for the team. A person then marks the report as a duplicate, not a product issue, or fixed, and you are told when it is fixed.

The AI does not change the software because of a report. The automatic repair above starts from the website checks failing, not from what clinics write in.

What is not automatic

  • Fixing a bug someone reported: a person decides and does it.
  • Security patching without a person watching: not yet.
  • Fixing one clinic’s own data: that goes through support.

Common questions

What do you do when you have an engineering problem?
If a key page fails, the repair helper puts the last working version back live within minutes, an AI writes a fix, other AI reviewers and the automatic tests check it, and then it goes live with a report emailed to the CEO. If it cannot find a safe fix it stops and says so. A person handles anything outside that.
What kind of engineering team do you have?
A small company. AI helpers do the around-the-clock watching and the first repair, and people set their rules and are told what they did. I would rather say that plainly than describe a large team that is not there.
How long does it take to fix a problem?
In the October 2026 tests, the working version was back live about three minutes after a deliberate break, and the fix was live within the hour. That is what was measured, not a guarantee. A problem the repair helper does not cover takes as long as a person needs.
Are security problems fixed automatically?
They are watched automatically, every hour, and the CEO is emailed. An AI patcher is switched on for same-version fixes, but it is new and still being proven, so a person makes sure each hole is fixed until it works end to end.
Does the AI fix the bugs I report?
No. An AI reads your report and writes a note for the team, and a person decides what to do. You are told when it is fixed.
Can the software go down?
Yes, any software can. Dr. Rosie checks itself every minute and repairs itself when a key page fails. For your own clinic’s internet or server outage, see the offline view article.