Launching today
kosa8
Secure sandboxes with an undo button for AI agents.
39 followers
Secure sandboxes with an undo button for AI agents.
39 followers
Kosa8 lets you run AI agents inside secure, isolated virtual machines without risking your own computer. Give Claude Code, Codex, or other agents their own sandbox, restrict what they can access, and checkpoint the entire running environment before risky changes. If something breaks, rewind in about a second — even running processes are restored. You can also fork environments to test multiple approaches in parallel, run local AI models, and manage everything from one dashboard.
@drew15 Great product
This is interesting. Keeping the API key on the host, with only a placeholder in the sandbox, is the bit I'd want most, since a real key sitting next to an unattended coding agent is the worrying part. Rewinding a bad run in about a second, processes and all, is a lovely touch too. Good luck today.
curious about that rewind in a second feature. feels like a way to save my skin when i inevitably mess something up. might be worth trying just for that.
Running each agent in its own microVM is like a safety net. Being able to make changes without accidentally blowing up something else sounds pretty neat. Best of luck with the launch