Skip to content
statesetSANDBOX
THE EXECUTION LAYER FOR AI

Give your agents
room to run.

A real environment for every next step. Give your agents isolated compute to run code, work with files, and turn a plan into something real.

See the code

$25 in trial credits No credit card required

ISOLATED BY DESIGNENV / 001
gVisor isolation
•••agent-workspaceILLUSTRATION

❯ python analysis.py

Artifact written to /workspace/report.csv

gVisor isolation
Code + shell execution
Files + artifacts
Kubernetes native
01 / FROM PROMPT TO PROCESS

Your agent has a plan.
Give it a computer.

Create a sandbox, run a command, collect the result. Familiar SDKs keep the infrastructure out of your agent loop.

  1. 01

    Create an environment

    Choose a runtime and set its resources.

  2. 02

    Let your agent work

    Execute code and read or write files.

  3. 03

    Keep the result. Close the sandbox.

    Collect output and release the compute.

Read the quickstart
$ npm install @stateset/sandbox-sdk
import { Sandbox } from '@stateset/sandbox-sdk';

const sandbox = await Sandbox.create({
  apiKey: process.env.STATESET_API_KEY,
  template: 'python',
});

try {
  const result = await sandbox.exec(
    'python -c "print(sum(range(100)))"'
  );
  console.log(result.stdout);
} finally {
  await sandbox.close();
}
OUTPUT4950Example result
agent.tsExplore the API
02 / BUILT FOR THE WORK

Beyond the chat window.

The tools your agents need to move from generating an answer to getting something done.

Coding agents

Give coding agents a workspace to install dependencies, edit files, and test their changes.

Build with the SDK

Data & analysis

Run Python against real files. Generate charts, transform datasets, and return useful artifacts.

Explore code execution

Browser workflows

Use browser-ready environments for automated research, UI testing, and web interactions.

Explore environments
03 / CONTROL, AT EVERY LAYER

Let agents explore.
Set the boundaries.

Keep execution separate from your application. Choose resource limits, scope access, and follow what your agents are doing.

Explore the architecture

Isolate execution

gVisor runtime support and non-root execution.

Define the environment

Explicit CPU, memory, and sandbox lifetime limits.

Control access

Scoped API keys, team roles, and encrypted secrets.

See what happened

Execution output, usage tracking, and audit logs.

04 / PRICING

$0.028/hour

4 vCPU · 8 GB sandbox, shared class.

See plans and rates

  • Billed per second

    Pay for the time a sandbox runs, not a rounded-up hour.

  • $25 trial ≈ 892 hours

    Of a 4 vCPU · 8 GB shared sandbox. No credit card required.

  • Dedicated when it can't be interrupted

    Guaranteed CPU and memory on on-demand capacity, never preempted.

YOUR NEXT BUILD STARTS HERE

Less setup.
More possibility.

Bring your model. Bring your code. We'll give it room to run.

Explore pricing