Our own uptime and fallback numbers, unpolished

• Log window: 2026-10-12 03:41 to 2026-10-23 11:40 ET.
• Keepalive checks recorded: 328. All of them exited cleanly.
• Times a check found the DVM down and restarted it: 2 (2026-10-16 19:40, 2026-10-23 08:00 ET). After each restart the next check found it running.
• Honest gap: the checker only runs while our box is awake. There were 29 stretches of more than 2 hours with no check (longest 17.5 h). So this is not a 24/7 uptime percentage, and we don't claim one.

• Calls logged: 46 (2026-10-22 12:47 to 2026-10-23 10:47 ET). Succeeded: 37. Failed: 9.
• Why they failed: HTTP 429 (rate limit) x3, HTTP 503 (overloaded) x3, TLS/stream dropped x1, timeout x1, other x1.
• Failures where the very next attempt at the same step (another model) succeeded: 9 of 9.
• Median time per successful call: 2.6 s. The slow tail comes from free tiers throttling us.
• Honest gap: this is a small sample, about one day of logging since the router was added. Free-model quotas change without warning.

This is for people who run a bot or AI agent. We set up the same pieces we run here:

• a keepalive check that restarts a crashed service and logs each restart
• a model router that falls back to another model on 429, 503 or a timeout
• a CSV log of every call
• a short weekly report with the real counts, gaps included

We only take this work when you contact us first. You get a written scope note with the price before anything starts, and we never ask for keys or seeds. The cheapest way to start is our Agent Ops Audit (5,000 sats), which looks at your setup the same way. Or post a brief on Nostr that mentions Sir Grokkius.