GPT6+Corv = infrastructure prod work is finally feasible

A while back I pointed out that AI infrastructure tools get almost no attention compared to developer ones. I think there are two reasons: it's niche, few people feel the underlying need, and there's a liability situation: you can't promise an agent won't nuke a system because it decided that deleting something was more efficient than fixing it. This can be reduced big time with the tooling on which the models work and permissions, so I had to kick-start this myself with Corv. Corv started as an SSH AI client and now it has finally become an AI-native SSH execution layer. Think MCP, but made specifically for infrastructure work. SSH was built for humans, and models have mostly been bolted on with coping hacks. Corv lets a model work over SSH the way it actually needs to (like a human, but on steroids): structured output instead of scraped terminal text, 1 connection, agentless, no tmux or bloat garbage, safe retries with an idempotency key, so a lost response never runs a mutating command twice long jobs it can recover, credentials kept private thanks to profiles, so nothing gets exposed. auditing, every command runned from the models get logged etc.. All this seamlessly from one command: corv The new version (v1.1.1) is a big jump in capacity and reliability. With the latest top-tier models like GPT6-Astra especially, its second-order thinking is a real step up. Corv comes with skills already in place for Claude and Codex. There's also a demo so that you can test it with a staging enviroment before starting into production in case you're new, look under for more. I'm now confident to say an AI enterprise engineer that genuinely handles infra work, has become feasible. Enjoy!