KyleVlaros

there are still a lot of things to do: AI fits infrastructure perfectly: Kubernetes and cloud are already API driven, so agents can observe real state, keep durable context, and take bounded action.

Avoid internet censorship, bypass Telegram and YouTube blocking, but do not default to a public VPN. Cheap VPS and open-source project with agent will do the trick.

The answer is absolutely not. The industry is optimizing the wrong layer. A general multimodal LLM is not replacing a radiologist any time soon, and even a domain-trained model is not a substitute for clinical responsibility.

Cool, thanks I will check it.

Portal is a tool I originally built for myself because existing AI coding tools felt too local and too chat-centric for real terminal and remote work.

Chat and CLI both flatten AI work into a linear stream. That is fine for single prompts or single commands, but not for real workflows where writing, execution, review, and decisions need to stay separate and reusable.

I wanted to share this partly because I assume other people are building similar notebook-style AI tools, and

I’d be interested to see them. There are so many AI products and side projects now that it is hard to keep up, so if you know something close to this model, send it over.

AI is changing operations fast by compressing triage, diagnosis, and decision prep into minutes instead of hours. But production is still not fully autonomous. Agents need reliable tooling, bounded permissions, clear runbooks, and approval gates. The winning model is semi-automatic: AI proposes, operators verify, systems execute with traceability. That balance delivers speed without surrendering control.

ktl is a Kubernetes CLI for building, previewing, and applying releases, but the real glue is its SQLite-backed local state. Helm and the broader Kubernetes workflow often feel fragmented, with too many separate tools for planning, rollout tracking, logs, and status, and ktl is basically an attempt to make that flow more cohesive. You still get fast feedback, rollout visibility, chart planning, and sandboxed build workflows, but SQLite is what keeps it all stitched together in practice. It holds run metadata, cached operational context, and UI/CLI state in one lightweight embedded store, so you get responsive behavior and useful history without adding more infrastructure.