I built my own AI agent, it's called the Pope Bot, and it works kind of like OpenClaw (Moltbot) but with two big differences: it's more secure and way more scalable. Here's the basic idea: You send the bot a task (via API, Telegram, whatever). The bot creates a branch in GitHub with that job. GitHub Actions picks it up, spins up a container, runs the task, and checks everything back into the repo when it's done. Nothing runs on your computer. It runs in the cloud. Your API keys? Stored securely in GitHub Secrets. They get injected at runtime and don't just sit around exposed somewhere. And like OpenClaw it can modify itself. I told it to add a scheduled task (a cron job) to run every 30 minutes. It figured out how to update its own config, pushed the changes back to the repo, and now it just... does that. On its own. It also builds its own tools over time. Say you tell it to check your email. It doesn't have a tool for that yet. So it looks at the API, figures out the call, writes the code, and saves that skill back into the repo. Next time? It already knows how. It only figures it out once. So your bot gets smarter the more you use it. And because everything lives in Git, if you ever wanted to share your assistant with someone, you just share the repo. They swap out their keys and they're exactly where you are. I'm working on getting this out to you all as soon as I can. It's one of those things that's easier to build than explain — but more coming shortly.