Sandbox
Secure, Isolated Code Execution for AI Agents
Code Without Consequences
Let AI Write Code. Keep Your Systems Safe.
AI agents that can code are transformative—but only when they can't break anything. Sandbox provides secure, isolated Docker environments where agents can write, test, and execute code without touching your production systems.
Complete Isolation
Each sandbox runs in its own Docker container with separate processes, filesystems, and network—completely isolated from your host.
GPU Acceleration
Optional NVIDIA GPU pass-through for ML workloads, data processing, and image generation with CUDA support.
Package Management
Install Python (pip) and Node.js (npm) packages on-demand. Packages persist within the sandbox session.
Web App Proxy
Serve web applications from your sandbox via secure HTTPS proxy. Perfect for Streamlit, Flask, and React apps.
Defense in Depth
Security Model
Multiple isolation layers protect your host system and other tenants. Sandboxes are ephemeral, resource-limited, and network-isolated by default—security isn't optional, it's foundational.
Container Isolation
Complete process and filesystem isolation ensures code running in one sandbox can never access another sandbox or the host system.
- Separate PID namespace per container
- Non-root user execution (uid 1000)
- Read-only root filesystem
- Writable /workspace only
Network Controls
Outbound internet access is disabled by default. When enabled, egress filtering controls what endpoints sandboxes can reach.
- No outbound connections by default
- No inter-container communication
- Inbound via secure proxy only
- DNS when outbound enabled
Pre-Configured Runtimes
Popular development environments ready to use, with on-demand package installation
Python 3.11+
NumPy, Pandas, Matplotlib pre-installed. Requests, BeautifulSoup for web scraping. Any pip package on-demand.
Node.js 20 LTS
Express, Axios pre-installed. TypeScript support. React/Vue/Angular builds. Any npm package on-demand.
GPU Runtime
CUDA 12.x and cuDNN pre-installed. PyTorch and TensorFlow with GPU support via pip.
Jupyter Kernel
Interactive notebook execution for data analysis, visualization, and iterative development.
The Tools Your AI Agents Need
Whether your agents are building data pipelines, web scrapers, machine learning models, or full-stack applications—Sandbox provides the runtime environment to make it happen without compromising your infrastructure security.
Global Infrastructure
Multi-Cloud Deployment
Deploy sandboxes across AWS, Google Cloud, or Azure. The platform automatically selects the optimal region based on user location for minimal latency.
Amazon Web Services
EC2 instances with EBS storage. Regions: us-east-1, us-west-2, eu-west-1, ap-southeast-1.
Google Cloud Platform
Compute Engine with Persistent Disk. Regions: us-central1, europe-west1, asia-east1.
Microsoft Azure
Virtual Machines with Managed Disks. Regions: eastus, westeurope, southeastasia.
Self-Hosted
Deploy on your own infrastructure for complete data sovereignty and compliance.
Web Application Proxy
Serve web apps from your sandbox through secure HTTPS proxy
Streamlit Dashboards
Build and serve interactive data dashboards. AI agents can create visualizations users can explore.
Flask / FastAPI
Deploy API backends and web services. Build custom endpoints for integrations and webhooks.
React / Vue / Angular
Build and serve full frontend applications. AI agents can create complete web experiences.
Gradio ML Interfaces
Create interactive machine learning demos. Upload images, adjust parameters, see results live.
From Code to Production. Instantly.
AI agents don't just write code—they deploy it. Start a web server in your sandbox, enable the proxy, and share a secure URL with anyone. Proxy URLs are unique per sandbox, support authentication, and expire automatically when the sandbox is destroyed.
Enable Coding Tools
AI Agent Sandbox Tools
When coding tools are enabled on an agent node, these sandbox capabilities become available—giving your AI the power to write, analyze, and execute code safely.
file_reader
Read files from the sandbox /workspace directory. Access code, data, and outputs.
file_writer
Write files to the sandbox. Create scripts, save outputs, build applications.
terminal
Execute shell commands. Run scripts, install packages, manage files.
code_analyzer
Analyze code for errors, security issues, and improvement opportunities.
Ready to Unleash AI Coding Power?
Give your AI agents the ability to write, test, and deploy code in secure, isolated environments. All the power of code execution, none of the risk to your infrastructure.