Hopx

Say an AI agent writes a Python script and you need to run it without trusting it. Hopx is a sandbox platform that spins up Linux micro-VMs for exactly that: running untrusted code and hosting AI agents.

Each VM is a Firecracker microVM, so code runs behind hardware-level security with kernel isolation.

Cold starts take about 100ms, which matters when an agent launches a fresh environment on every request.

Hopx does not cap execution time, so a job or agent can keep running for hours or days.

State persists across agents, notebooks and jobs, so you can pick work up where it stopped.

Compute scales up in milliseconds and drops to zero on its own when idle.

Billing is per second for compute, memory and storage, so you pay for what actually runs.

SDKs cover several popular programming languages, which keeps integration with an existing agent stack simple.

Free credits let you start without a credit card, a low-risk way to test the speed claims yourself.