v1.6.211: Tom AI can now FIX things, not just explain them (opt-in beta) + Tailscale/installer/VPN fixes
Posted by tomspark
Big update today. Tom AI agent beta (Backers/Legends, opt-in). Until now Tom AI could only tell you what's wrong. As of v1.6.211 it can look inside your box and act: check what's running, read your download queue and logs, and propose the fix itself — like restarting a stuck app or repairing media-folder permissions. Every action shows you a Run it / Skip button first. Nothing happens without your click, and after it acts it re-checks your box and rolls back if anything got worse. The hard limits, because they matter: the agent has a fixed menu of safe tools — no shell, no free-form commands. It cannot install apps, add sources, search for content, or download anything. Those tools don't exist for it, period. It's a server mechanic, not a content butler. And it can't touch the VPN kill-switch or the dashboard itself. It's OFF by default. To try it: update to v1.6.211 (sparkbox upgrade or the dashboard update button), then Settings → AI Troubleshooting → enable the agent beta. You'll get a consent prompt that spells everything out. It's a beta — if it does anything weird, post here and I'll see it. Also in v1.6.211: - Installing on a NAS without sudo now stops upfront with the exact command to re-run, instead of "succeeding" into a broken half-install (no sparkbox command, apps that won't start). - Tailscale: custom Funnel/serve setups no longer reset overnight. Declare them once in modules/tailscale/serve.user.json and they survive every update and restart — full example at tomsparkbox.com/docs.htmltailscale-custom-serve - Surfshark over OpenVPN no longer crash-loops (their servers push a Windows-only DNS option that broke Linux; we now ignore it). Full notes: tomsparkbox.com/changelog.html