Skyportal watches your deployments, Kubernetes events, GPU metrics and logs, and explains what changed when production breaks. It proposes a fix, waits for your yes, and verifies it with fresh read only evidence. MIT, pip install skyportalai.
Hey PH. We build Skyportal, an open source AI infrastructure engineer. It watches your deployments, Kubernetes events, GPU metrics and logs, builds a timeline of what changed, and when production breaks it explains the change that did it, ships the fix as a PR, and checks that the fix held. The part we care most about is the gate: every state changing command stops at a prompt for your yes, runs on the hosts you named, and its own claim of success counts for nothing, a plan only closes on fresh read only evidence. MIT licensed, Python, pip install skyportalai. Docs at docs.skyportal.ai. Tell us where you would loosen the gate, and where it is still too loose.