Sandboxes in the cloud
Run apps, databases, and background tasks. Use the tools you already know.
Couldn't load this page.
Get your agents working fast. Let them build in sandboxes and deploy persistent apps and databases.
npm install -g @lizard-build/cliPay as you go. No monthly subscription.
Sessions run back to back for a full month (730 hours).
10 sandboxes × 730 hours × $0.009/h
Sandbox egress is included up to 2 TB per account over a rolling 30 days. Read the limits
Run code, keep files, preview apps, and monitor usage.
Run apps, databases, and background tasks. Use the tools you already know.
Keep files between runs. Mount storage at /workspace.
Launch in US East or EU West. Choose where your agent runs.
Use Lizard CLI in your terminal or Lizard SDK in your code.
Preview your frontend over HTTPS. Share a link while your sandbox runs.
CPU and memory metrics. Logs and command output.
Lowest median. Lowest average. Measured against E2B, Daytona, and Modal.
Lowest median time
| Provider | Median | Average | p90 | p95 | p99 | Maximum |
|---|---|---|---|---|---|---|
| Lizard | 196 ms — Lowest time | 238 ms — Lowest time | 323 ms — Lowest time | 522 ms — Lowest time | 1,078 ms | 6,054 ms |
| E2B | 282 ms | 336 ms | 472 ms | 590 ms | 889 ms — Lowest time | 1,687 ms — Lowest time |
| Modal | 972 ms | 1,000 ms | 1,208 ms | 1,318 ms | 1,640 ms | 2,119 ms |
| Daytona | 977 ms | 1,063 ms | 1,436 ms | 1,611 ms | 2,547 ms | 6,441 ms |
300 runs per provider. Median is the middle result; p90 means 90% of runs finished within that time. Rankings apply to this test.
An AI sandbox is a separate place for an agent to run code, use tools, and work with files. Lizard Sandboxes let agents build, test, and debug in the cloud without running that code on your computer.
Sign in to Lizard, choose a project, and create a sandbox from a ready-made template. Choose base for shell tasks, codex for coding tasks, or interpreter for Python, then select a size and region.
Yes. Run Python scripts with the interpreter template, or use shell commands to run tools and tests. Read command output and save results to files. Each command starts a new process.
Lizard uses a pool of prepared Sandboxes to reduce setup time. Startup time depends on the template, size, region, and available capacity.
Each sandbox runs in its own Kubernetes pod using runc and shares the host kernel. It uses container isolation, without a separate virtual machine kernel. Give the agent only the credentials and access needed for its task.
You choose the lifetime when creating a sandbox, including no automatic expiry for longer tasks. A sandbox stops at its deadline even if a command is running. Commands do not extend the deadline, and idle time does not stop it early.
Yes. Attach Persistent Volumes at /workspace to keep files after a sandbox stops and reuse them in later runs. A volume connects to one sandbox at a time. Storage stays on one node, so export files you need to protect against node loss.
Yes. With access to your project, an agent can deploy apps and databases through Lizard CLI. They run as separate resources on Lizard and keep running after you stop or delete the sandbox.
Yes. Create a separate sandbox for each task so agents can run commands and tests at the same time. The number of Sandboxes you can run depends on your account limits and available capacity in the region.
The Small size costs $0.009/hour for 2 vCPU and 4 GB RAM. Billing runs by the second, including idle time, with CPU, RAM, and sandbox egress included. Persistent Volumes cost extra. Older sandboxes without a size keep their original usage-based billing.
Trial or promo credits, when available in your account, can cover Sandboxes usage. Standard accounts use pay as you go with no monthly subscription. Trial and promo credits can expire; purchased credits do not.
You don't need a platform team to go live. Your whole cloud, one CLI command away.
We use cookies for essential site functionality and analytics. See our Cookie Policy.