Agent Sandbox

Image

Prerequisites

• A running Kubernetes cluster with the Agent Sandbox Controller installed.
• The Sandbox Router deployed in your cluster.
• A SandboxTemplate named python-sandbox-template applied to your cluster. See the Python Runtime Sandbox guide for setup instructions.
• The Python SDK installed: pip install k8s-agent-sandbox.

Architecture

The Agent Sandbox architecture separates the Template (the definition) from the Claim (your Python request) for a very specific reason: speed. When a user applies a SandboxTemplate to a Kubernetes cluster, the controller typically spins up a SandboxWarmPool. These are pre-initialized, running pods that have already pulled your specific Docker image. When a Python script calls client.create_sandbox("sandbox-template"), it instantly grabs one of these pre-warmed pods.

Workarounds

1. Install dependencies via sandbox.commands.run() function

from k8s_agent_sandbox import SandboxClient
client = SandboxClient()
sandbox = client.create_sandbox("python-sandbox-template")
# Dynamically install a package before running your main logic
sandbox.commands.run("pip install custom-package==1.0.0")
response = sandbox.commands.run("python -c 'import custom_package; print(\"Success!\")'")

[NOT FOR PUBLISHING] 2. Use kubernetes Python SDK to create a new SandboxTemplate instance

We can create a Python script that uses kubernetes SDK to deploy a custom template, or update an existing template with our custom Docker image.