I built Aegil, an autonomous Site Reliability Engineering (SRE) agent for Kubernetes with deterministic log analysis. It connects to your cluster and log store, watches the state, diagnoses by facts, and fixes infrastructure problems, while you just talk to it in plain language. The key point: the language model acts through real tools but within deterministic bounds. The danger class of every command is decided by a pure function outside the model, anything irreversible needs operator confirmation, and every execution goes through guards and an audit.

I did not just make a promo page, I took the system apart: how confidence is computed with Bayesian odds, why the model physically cannot wipe your data, how the privileged node-agent works without a backdoor, and where it all has its limits.

Promo page: https://maxbirkin.com/aegil The series (12 parts): https://maxbirkin.com/blog/system-overview

And yes, I refreshed the site too )