zeishv0.1 Public Beta will be live soon

Git-like cloning for
MicroVM sandboxes

Ephemeral MicroVM sandboxes with instant boot from snapshots. Clone, branch, and snapshot environments — built for reproducible AI agent runs.

Get Started Visit the docs
SDKs & APIs

Drop it into your stack

First-class SDKs for every language your agents run in — or just hit the HTTP API directly.

1import { zeish } from "@zeish/computesdk-provider";
2
3const compute = zeish({ apiKey: process.env.ZEISH_API_KEY! });
4
5const { sandbox } = await compute.sandbox.create({ name: "agent-run" });
6
7const result = await compute.sandbox.runCommand(sandbox, "python agent.py");
8
9console.log(result.stdout);
~0msBoot time
0%Reproducibility
—Cold Starts
Sanbox Orchestration

Secure Sandboxes for
AI Agents

Safe, isolated environments for executing untrusted code, running evals, and tool use. Clone entire environments in milliseconds.

auto-gpt-runner

running•agent-sandbox-01
Uptime
99.97%
vCPUs
2
Memory
2.1 GB
Boot Time
124ms

Network Activity

Global Edge Network

Deploy instantly to 3 regions worldwide.

Operational
Offline

Network Activity

Live
vm.clone47ms
US East12:45:23
vm.create89ms
EU West12:45:22
snapshot.save156ms
AP South12:45:21
vm.boot198ms
US East12:45:20

How it works

Create a base VM, fork it at any point, and branch infinitely — all with instant boot times.

Base VMubuntu:22.04Fork Aagent-evalFork A1exp-variant-1Fork A2exp-variant-2Fork Bcode-interpreter
Base VMForked VMSnapshot
Tip: Right-click a node to clone, snapshot, suspend, or destroy

Step 01

Initialize

Create a base MicroVM with your preferred OS and dependencies. The foundation state.

$zeish init base

Step 02

Fork

Branch from any snapshot point. Each fork is a fully isolated environment linked to the parent via copy-on-write.

$zeish clone base@clean --name agent-run

Step 03

Scale

Run parallel experiments. Merge results or discard environments — all ephemeral.

$zeish clone agent-run@checkpoint --name experiment-2
Audit Logs

Every action, recorded.

Comprehensive audit trail for every VM operation, snapshot, clone, and API call. Full compliance visibility out of the box.

vm.create

Created MicroVM 'eval-runner-a7f'

agent-worker-01•10.0.1.42•14:23:41
snapshot.save

Snapshot 'checkpoint-1' saved

system•10.0.1.10•14:23:39
clone.start

Clone from 'base@v2.1' initiated

ci-pipeline•10.0.2.15•14:23:37
vm.exec

Executed 'python eval.py'

agent-worker-03•10.0.1.88•14:23:35
vm.destroy

MicroVM 'temp-sandbox-f2a' destroyed

system•10.0.1.10•14:23:33
api_key.create

New API key created for 'prod'

admin•192.168.1.50•14:23:30

Use Cases

How teams use Zeish to build reliable AI infrastructure.

Evals

Agent Evals

Run dozens of evaluation scenarios in parallel, each in a clean MicroVM clone. Compare outputs deterministically.

Experiments

Parallel Experiments

Clone a baseline and test multiple prompts, tools, or model versions simultaneously across isolated environments.

Security

Secure Tool Execution

Let agents call shell commands, install packages, or write to disk — all inside an ephemeral sandbox that gets destroyed after.

CI/CD

CI for Agents

Every commit triggers an agent run inside a cloned MicroVM. Reproducible environments for every test in your pipeline.

Data

Dataset Pinning

Snapshot a VM with a specific dataset loaded. Clone from that snapshot to guarantee identical data across all runs.

Runtimes

Sandboxed Interpreters

Safe code execution for LLMs. python-interpreter, node-runner, or any custom runtime can be spun up in milliseconds.

Technical FAQ

Common questions about Zeish MicroVMs.