Agent Sandbox
Image
Create a Sandbox with custom dependencies.
Prerequisites
- A running Kubernetes cluster with the Agent Sandbox Controller installed.
- The Sandbox Router deployed in your cluster.
- A
SandboxTemplatenamedpython-sandbox-templateapplied 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!\")'")