Platform for AI-generated code.

Cost-effective infrastructure for running AI code, with sub-100ms cold starts and thousands of sandboxes in parallel.

$pip install lizard-sdk
from lizard import Sandbox
sandbox = Sandbox.create("code-interpreter-v1", project_id="proj_abc123")
result = sandbox.process.exec_("python -c 'print(2 ** 10)'")
print(result.stdout) # 1024
sandbox.kill()

Fast, isolated, stateful.Infrastructure built for AI agents.

90ms cold boot

Lightning-fast

Firecracker microVMs boot in milliseconds, not minutes. Spin one up per request and tear it down when the work is done.

Isolated runtime

Every sandbox is a separate VM with its own kernel. Run untrusted, AI-generated code with zero risk to your infrastructure.

Massive parallelization

Fan out thousands of concurrent sandboxes for evals, batch jobs, or agent swarms — each fully isolated from the rest.

Programmatic control.Execute, files, snapshots, and code.

exec.py
# Run a command, get stdout, stderr, exit code
result = sandbox.process.exec_("npm test", workdir="/app")
print(result.stdout, result.exit_code)

More than a sandbox.The runtime agents actually need.

node:22templatesandboxes

Snapshots & templates

Pre-bake dependencies into a template, then launch identical sandboxes instantly.

us-easteu-west

Regional deployment

Place sandboxes close to your users or data across us-east and eu-west.

disk + memoryexecexectimeout

Stateful by design

Sandboxes keep disk and memory between calls, then clean themselves up on timeout.

shared volume

Volumes

Mount shared, persistent storage across many isolated sandboxes at once.

How sandboxes stack up.See it for yourself.

Compared bylizard
Starting price
Free tier
Cold boot time
Isolation model
Max concurrent sandboxes
Persistence (snapshots, templates, volumes)
Regions
SDK languages
Programmatic control (exec, fs, snapshots, code interpreter)
Billed while idle
Support

Frequently asked questions

A Lizard sandbox is an isolated Firecracker microVM you create, control, and tear down through the API or SDK — its own filesystem, process tree, and kernel, built for running AI-generated or otherwise untrusted code.

Every sandbox runs in its own Firecracker microVM with its own kernel, not a shared container. Code inside one sandbox can't see or affect another sandbox, or your production infrastructure.

Cold boot is milliseconds, not seconds — fast enough to spin one up per request and tear it down when the work is done, without keeping a warm pool around.

Official SDKs are available for Python and TypeScript, with more languages on the way.

Yes — sandboxes are stateful by default, keeping disk and memory between calls. You can also snapshot a sandbox into a template to launch identical copies instantly.

Yes — place a sandbox close to your users or data. Currently available in us-east and eu-west, with more regions on the way.

You pay only for the seconds a sandbox actually uses CPU and memory — not for the whole time it stays open, the way most providers charge. Idle sandboxes shut themselves down, so nothing quietly runs up a bill in the background. That makes Lizard one of the cheapest sandboxes on the market.

Build with AI. Ship with Lizard.

You don't need a platform team to go live. Your whole cloud, one CLI command away.

Try for free

No credit card required

We use cookies for essential site functionality and analytics. See our Cookie Policy.