Sandbox Integration Guide

Choose a provider based on your needs: E2B for fast stateless code interpretation, Daytona for full development environments with VNC preview, or Docker for self-hosted control. Then implement health checks, resource limits, and network isolation as shown in the integration patterns.

Use E2B for quick Python/JavaScript execution with sub-second startup, Daytona when you need full Linux environments with multi-service orchestration and VNC preview, and Docker when you require self-hosted infrastructure, custom configurations, or high-volume execution.

Combine the Daytona workspace state with a service health poll of the sandbox /health endpoint. Derive a unified status: started plus healthy services means LIVE, degraded services means FAILED, and stopped or archived means OFFLINE.

Use sandbox.files.write to upload content to paths like /workspace/data.csv before executing code, then use sandbox.files.read to retrieve generated outputs. The code interpreter can access these files during notebook cell execution.

Apply defense in depth: drop all Linux capabilities, enable no-new-privileges, mount a read-only root filesystem with tmpfs for /tmp, use an internal bridge network to block internet access, and enforce CPU, memory, and execution timeout limits.

E2B focuses on stateless code interpreter sessions, so it is less suited for long-running processes, multi-service applications, or workloads needing persistent full development environments. For those cases, Daytona or self-hosted Docker is a better fit.