Hey Mike, a crash loop like that on a DigitalOcean droplet right after adding memory and team members is almost always RAM. Right after a crash, run free -h and dmesg | grep -i oom. If the OOM killer shows up, that is it: multiple agents plus the memory embeddings blow past a 1-2GB box fast. Quickest fix is to strip back to one agent, drop the custom memory tweaks, and confirm it stays up for a full day, then add team members back one at a time so you can see what actually breaks it. The part where it cannot fix itself is usually just the gateway dying with nothing to restart it, so put it under a process manager or a small cron health check that restarts it when it is down. If you paste the actual error from the logs around a crash, happy to help you pin it down.