Tom AI now learns from 150+ real boxes — and waits longer to judge fixes
Posted by tomspark
1.6.497 ships with fleet-tested fix recipes. When Tom AI spots a problem, it now proposes the exact steps that worked on boxes like yours, with real success counts — '150 of 207 boxes matching your setup' — instead of generic advice. I've been chasing this since we started collecting anonymous fix outcomes; the data finally got dense enough to be useful. I also fixed the impatience problem. Tom AI was declaring fixes failed after 2.5 minutes, which broke anything legitimate — media apps on a NAS genuinely need 5 minutes to boot. Now it respects each app's own startup window, so a working fix actually gets reported as working. Three more things in this release: - Your box downloads a public known-issues list daily and warns you (banner + phone push) when a problem you haven't hit yet is already fixed in a newer version — all checked locally, nothing leaves your box. - Boxes sharing fix outcomes now also report the problem type, which is what makes those success numbers honest and useful. - Fixed the password field eating 'wireguard' and other app names as literal passwords — recognized them as app references instead. Head to the Updates tab on your dashboard to pull 1.6.497. If something breaks, post it here in d/sparkbox. --- Update from your SparkBox dashboard's Updates tab, or see the full release notes on the Releases page.