Helm repository for Kubernetes Agent Sandbox (sigs.k8s.io/agent-sandbox).
This repository packages and maintains Helm charts to deploy and manage isolated execution sandboxes for AI agents, LLMs, and code execution workloads on Kubernetes.
| Chart | Version | Description | Artifact Hub |
|---|---|---|---|
agent-sandbox |
All-in-one chart bundling Controller, Router, Python/Bash runtime, WarmPools, NetworkPolicy, and RBAC | View on Artifact Hub |
helm repo add dovakiin0 https://dovakiin0.github.io/agent-sandbox-helm/
helm repo update
helm install agent-sandbox dovakiin0/agent-sandbox \
--namespace agent-sandbox-system \
--create-namespace
Check that the Controller, Router, and WarmPool are running:
kubectl get pods -n agent-sandbox-system
kubectl get sandboxtemplates,sandboxwarmpools,sandboxes -n agent-sandbox-system
+------------------------------------+
| Client / AI Orchestrator |
+-----------------+------------------+
|
+---------------------+--------------------+
| HTTP / WebSocket | Sandbox CRD / Claim
v v
+-----------------------+ +-----------------------+
| Sandbox Router | | Agent Sandbox Ctrl |
| (High-throughput | | (Reconciles Sandbox, |
| Reverse Proxy) | | WarmPools, Templates)|
+-----------+-----------+ +-----------+-----------+
| |
| |
v v
+------------------------------------------------------------------+
| Isolated Sandbox Pods |
| - automountServiceAccountToken: false |
| - Non-root (UID 1000) / Dropped Capabilities |
| - gVisor / Kata Isolation Support |
| - Strict NetworkPolicy (Blocks 169.254.169.254 & RFC1918) |
+------------------------------------------------------------------+
See the detailed configuration guide and parameters list in the Chart README.
Example custom configuration (values.yaml):
# Enable gVisor VM/kernel isolation
security:
runtimeClassName: "gvisor"
# Scale pre-warmed pool
warmPool:
replicas: 5
# Customize sandbox resources
sandboxTemplate:
resources:
requests:
cpu: "500m"
memory: "1Gi"
limits:
cpu: "2000m"
memory: "2Gi"
Apply customization:
helm upgrade --install agent-sandbox dovakiin0/agent-sandbox \
-f values.yaml \
--namespace agent-sandbox-system
agent-sandbox-helm/
├── .github/
│ └── workflows/
│ └── release.yaml # Automated chart release & GitHub Pages workflow
├── charts/
│ └── agent-sandbox/ # Helm chart
│ ├── Chart.yaml # Chart metadata and Artifact Hub annotations
│ ├── README.md # Artifact Hub package README
│ ├── values.yaml # Default values
│ ├── values.schema.json # Values JSON schema
│ ├── crds/ # Agent Sandbox Custom Resource Definitions
│ │ ├── sandboxes.agents.x-k8s.io.yaml
│ │ ├── sandboxclaims.extensions.agents.x-k8s.io.yaml
│ │ ├── sandboxtemplates.extensions.agents.x-k8s.io.yaml
│ │ └── sandboxwarmpools.extensions.agents.x-k8s.io.yaml
│ └── templates/ # Kubernetes resource templates
│ ├── _helpers.tpl
│ ├── controller-deployment.yaml
│ ├── controller-rbac.yaml
│ ├── router.yaml
│ ├── sandbox-template.yaml
│ ├── sandbox-warmpool.yaml
│ ├── networkpolicy.yaml
│ └── backend-rbac.yaml
├── artifacthub-repo.yml # Artifact Hub repository verification
└── README.md
This repository uses Helm Chart Releaser to automatically publish new chart versions:
version in charts/agent-sandbox/Chart.yaml.main or master:
git add .
git commit -m "Release agent-sandbox v1.0.3"
git push origin master
gh-pages branch.This project is licensed under the Apache 2.0 License.
Upstream project: Kubernetes SIGs Agent Sandbox.