How we built a sub-100ms cold start Cloud for AI Agents

After years of building ephemeral environments at Bunnyshell; helping teams spin up full-stack environments per pull request; we took everything we learned about isolation, speed, and reliability and applied it to AI agents.

That’s how Hopx was born — a Firecracker-based cloud that launches fully isolated Linux micro-VMs in under 100 ms, with persistent storage, WebSocket streaming, and pay-per-second billing.

You can now run AI agents, notebooks, or long-running workflows directly via API or SDKs (Python, JS, Go, .NET, Java, PHP).
Each agent runs in its own VM — safely executing LLM-generated or user-submitted code with full PTY and file access, no host exposure.

We’d love your feedback, how would you use this kind of cloud for your agents?