Evals
Agent Evals
Run dozens of evaluation scenarios in parallel, each in a clean MicroVM clone.

Git-like cloning for MicroVM sandboxes
Hardware-isolated MicroVMs with instant boot from snapshots. Snapshot, branch, and fork agent environments the way you'd clone a repo.
npm i @zeish/computesdk-providerTrusted for high-performance agent workloads
Cloning
Create a sandbox, snapshot it, and fan out parallel clones for agent runs, copy-on-write all the way down. Right-click any node in the canvas to clone, snapshot, suspend, or destroy.
Parent pages stay shared. Each fork only pays for its writes.
Security
Dedicated MicroVMs, snapshot boot, and serverless-style APIs, so shell, network, and disk stay inside a boundary you control, not a shared container kernel.
Fleet
Every sandbox lands on a physical node running Firecracker or Cloud Hypervisor, its own kernel and hardware-enforced boundary, not a namespace on a kernel your neighbors also touch. The control plane schedules across the fleet; you never see the rack.
Use cases
Six workload patterns that need isolated, reproducible agent environments.
Evals
Run dozens of evaluation scenarios in parallel, each in a clean MicroVM clone.
Experiments
Clone a baseline and test multiple prompts, tools, or model versions simultaneously.
Security
Let agents call shell commands and write to disk inside an ephemeral sandbox.
CI/CD
Every commit triggers an agent run inside a cloned MicroVM, reproducible every time.
Data
Snapshot a VM with a dataset loaded. Clone from that snapshot for identical data.
Runtimes
Safe code execution for LLMs. Python, Node, or any custom runtime in milliseconds.