BLAZING-FAST SANDBOXES FOR
|

Sandboxes are ideal for AI agents that execute code, interactive development environments, data analysis, and CI/CD workflows.

✓ ~100ms startup time
✓ fully isolated Linux environment
✓ unlimited session time

Install the Bunnyshell SDK

1npm install @bunnyshell/sdk

Create and use a sandbox

1import { Bunnyshell } from '@bunnyshell/sdk'
2
3const bunnyshell = new Bunnyshell()
4const sandbox = await bunnyshell.create({ language: 'typescript' })
5
6const response = await sandbox.process.codeRun('console.log("Hello!")')
7console.log(response.result)

Isolated at the VM Level

Firecracker microVMs provide hardware-level security and kernel isolation — far beyond containers or serverless functions.

Startup in ~100ms

Sandboxes launch from prebuilt snapshots, allowing near-instant cold starts instead of seconds or minutes.

Run Continuously

No execution time limits. Keep agents, notebooks, or jobs running for hours, days, or weeks — with full state persistence.

Built for AI Execution

Native support for code execution, file operations, background jobs, desktop automation, MCP servers, and agent-driven workflows — all exposed through a programmable API.

THE FASTEST, MOST FLEXIBLE RUNTIME FOR AUTONOMOUS AI

Forkable MicroVM Snapshots

Create, clone, fork, pause, and resume full VM environments in ~100ms — enabling branching agent workflows, parallel execution, and instant recovery.

68ms

84ms

73ms

91ms

93ms

Global Low-Latency Runtime

Run sandboxes close to your models or users, across multiple regions worldwide — optimized for real-time AI execution.

Programmatic Control

Full API access for agents to write code, manage files, run commands, and continuously evolve their own environment.

Simple, Powerful SDK

npm install @bunnyshell/sdk

Connect to sandboxes with clean APIs in your favorite language

Execute Code

Run Python, JavaScript, and more with rich output capture

File Operations

Upload, download, watch files with full filesystem access

Process Control

Start, monitor, and manage long-running background processes

Port Forwarding

Expose services and access them publicly

Snapshots & Resume

Save and restore sandbox state instantly

execute-code.js

1const { Bunnyshell } = require('@bunnyshell/sdk');
2
3const client = new Bunnyshell({ apiKey: process.env.API_KEY });
4const sandbox = await client.sandboxes.create({ template: 'python' });
5
6// Execute Python code with matplotlib
7const result = await sandbox.execute(`
8import matplotlib.pyplot as plt
9import numpy as np
10
11x = np.linspace(0, 10, 100)
12plt.plot(x, np.sin(x), label='sin(x)')
13plt.plot(x, np.cos(x), label='cos(x)')
14plt.legend()
15plt.title('Trigonometric Functions')
16plt.savefig('/workspace/plot.png')
17print('Chart generated!')
18`, { language: 'python', captureRichOutput: true });
19
20console.log(result.stdout);  // "Chart generated!"
21console.log(result.images);  // Array of base64 PNG images

Full desktops that agents
can control with code.

Secure, isolated virtual desktops powered by Firecracker microVMs — designed for AI agents that need full GUI interaction, browser automation, visual reasoning, or desktop software control. Access via VNC or browser (noVNC), run real desktop apps, and automate everything programmatically.

AGENT_19HT50

AGENT_21H41C

AGENT_8F3K2A

AGENT_4D7M9P

AGENT_6N2L8R

Full Linux desktop with root access, Firefox, Chrome, VS Code, terminal, and GUI apps.

Perfect for:

• Autonomous AI agents with browser automation
• Visual testing & GUI-driven workflows
• Desktop application reasoning
• Secure remote development environments

Includes:

• XFCE desktop
• VNC + browser access
• Chrome, Firefox, VS Code, Gimp, Office tools
• Full filesystem and process control via SDK

Linux (Ubuntu)

Linux desktop with full root access. Ready for automation, development, and testing with complete programmatic control.

MacOS

macOS sandboxes with full GUI for iOS/mobile app testing. Ideal for iOS development agents, Xcode automation, and vision-based AI workflows on macOS.

Windows

Full Windows desktop with programmatic control. Designed for Windows-specific testing and desktop automation, enterprise workflows, and RPA (Robotic Process Automation) powered by AI agents.

/ USE CASES

BUILT FOR AI AGENTS, LLM EXECUTION, AND SECURE MCP INFRASTRUCTURE

Run Untrusted Code

Safely execute user-submitted or LLM-generated code in isolated microVMs.

■~100ms cold start (Firecracker microVMs)

■Full Linux with file/exec/PTY access, no host exposure

■Simple APIs: execute, execute/rich, commands/stream

Run AI Agents

Launch agents that write and execute code in dedicated runtimes.

■Multi-language execution + real-time WebSocket streaming

■Persistent state via IPython kernel and filesystem

■Per-agent isolation with controllable background processes

Run Data Analysis Notebooks

Spin up Jupyter notebooks with ML libraries preinstalled.

■Persistent IPython + rich outputs (matplotlib/plotly/pandas)

■File upload/download and result caching

■Easy port exposure to access Jupyter securely

Run Deep Research Agents

Autonomous agents that iterate, gather data, and analyze continuously.

■Long-running processes with no arbitrary timeouts

■File watching, logging, and system metrics endpoints

■Fast relaunch/snapshot flows (~100ms)

Run Autonomous AI Workflows

Multi-step pipelines that need code execution and tools.

■execute/background + API-level orchestration

■Result caching and idempotent restarts

■Fine-grained resource/timeout controls per step

Run Computer Use Agents

Agents that control a full cloud desktop (GUI, browser, apps).

■Desktop automation: mouse, keyboard, windows, clipboard

■VNC/noVNC streaming, screenshots, and screen recording

■Optional browser automation (Firefox/Chromium)

Run Background Automations

Workers, schedulers, and recurring jobs that must keep running.

■Background processes with list/kill/status APIs

■Live logs via WebSocket; health & metrics endpoints

■Safe self-update to restart/upgrade the agent

Run Reinforcement Learning

Train/evaluate RL agents in isolated, repeatable environments.

■Continuous execution with on-disk persistence

■Stream metrics, videos, and artifacts in real time

■Strong isolation for reproducible experiments

Run Secure MCP Servers

Host MCP servers/tools in a controlled, isolated perimeter.

■Per-server network and filesystem isolation

■Safe port exposure for tools and endpoints

■Rapid startup (~100ms) with persistent sessions

Run Long-Running Jobs

Jobs that need hours or days without forced shutdowns.

■No artificial runtime limits

■Process management + large artifact downloads

■VM-level stability with near-native performance

Deploy your first agent runtime today

Whether you're building agents, MCP servers, or LLM infrastructure, Bunnyshell gives you the runtime. Firecracker-powered sandboxes with full control, zero lock-in in under 100 ms.

Frequently asked questions

Everything you need to know about Bunnyshell Sandboxes.

/ QUESTION

/ CATEGORY

What is a Bunnyshell Sandbox?

Getting Started

How is it different from containers or serverless functions?

Getting Started

How fast do sandboxes start, and how long can they run?

Performance & Runtime

Can I safely run untrusted or AI-generated code?

Security & Compliance

What operating systems are supported?

Platform & Templates

What templates can I use?

Platform & Templates

How do I interact with a sandbox?

Integration & API

What languages are supported?

Platform & Templates

Is Bunnyshell secure and enterprise compliant?

Security & Compliance

Can I deploy sandboxes in different regions or on-premises?

Security & Compliance

How is pricing calculated?

Pricing

Are sandboxes suitable for MCP servers and AI agent infrastructure?

Integration & API

Do I need a credit card to get started?

Getting Started

Can I snapshot or clone a sandbox for parallel execution?

Performance & Runtime

Is this suitable for CI/CD and automated testing?

Integration & API

What happens if a sandbox fails or shuts down unexpectedly?

Getting Started

BLAZING-FAST SANDBOXES FOR|

Isolated at the VM Level

Startup in ~100ms

Run Continuously

Built for AI Execution

THE FASTEST, MOST FLEXIBLE RUNTIME FOR AUTONOMOUS AI

Forkable MicroVM Snapshots

Global Low-Latency Runtime

Programmatic Control

Simple, Powerful SDK

Execute Code

File Operations

Process Control

Port Forwarding

Snapshots & Resume

Full desktops that agentscan control with code.

Linux (Ubuntu)

MacOS

Windows

BUILT FOR AI AGENTS, LLM EXECUTION, AND SECURE MCP INFRASTRUCTURE

Run Untrusted Code

Run AI Agents

Run Data Analysis Notebooks

Run Deep Research Agents

Run Autonomous AI Workflows

Run Computer Use Agents

Run Background Automations

Run Reinforcement Learning

Run Secure MCP Servers

Run Long-Running Jobs

Deploy your first agent runtime today

Frequently asked questions

BLAZING-FAST SANDBOXES FOR
|

Full desktops that agents
can control with code.