The Weekly Upkeep a Private AI Needs
Short answer: Every week: update the models and the harness when new versions ship, respond to host bans and price changes, rotate API tokens and keys, check that backups restore, and read the logs for errors. It runs for as long as you keep the system.
The list, as the pillar page gives it
Update the models and the harness when new versions ship. Respond to host bans and price changes. Rotate API tokens and keys. Check that backups restore. Read the logs for errors.
Each item is small on its own. Together they are the difference between a system that still works in six months and one that drifted out of date — the pillar page's warning is that if you stop for a month, it drifts out of date.
Where the weekly hours come from
The upkeep is not one job; it is the weekly tail of every step. The host list keeps moving. Paying hosts takes time each week. New model releases need testing before they replace the one that works. Backup checks and key rotation belong to failover. The interface and harness need attention when the pieces under them update.
Added up, the weekly hours in the six steps come to about 7 to 14 hours a week — the pillar page's estimate, not measured data — and that runs for as long as you keep the system.
Upkeep is also where breakage gets caught
Reading the logs and checking that backups restore are the two items that find problems while they are still small. A backup that no longer restores, found on a quiet weekday, is an errand. The same discovery during a host shutdown is the emergency described in the failover page.
Related DIY pages
Back to the pillar: What it takes to do this yourself
Adults only (18+). What it takes to do this yourself · Apply