Kubernetes-native sandbox for running untrusted AI-agent code on GKE, using gVisor kernel isolation, with pod snapshots for pause and resume and high-rate sandbox creation. Built on the open-source Kubernetes Agent Sandbox project.
FluffyStack toolsJurisdictional exposure
Attributes
- Ga Year
- 2026
Sub-services (3)
Agent Sandbox CRDs
Custom Resource Definitions for sandbox lifecycle, per-agent isolation policies
Pod Snapshots
Save and restore sandbox state across agent task executions
Axion N4A Optimisation
ARM-native execution profile tuned for agent workloads
Compliance & Certifications
This service is attested for the following frameworks. Always verify with the provider before relying on a specific compliance posture.
Where this runs
Sovereign regions (1)
- T-Systems Sovereign Cloud · FrankfurtT-Systems Sovereign Cloud powered by Google Cloud
Commercial regions (43)
Europe (13)
- Belgium
- Finland
- Paris
- Berlin
- Frankfurt
- Milan
- Turin
- Netherlands
- Warsaw
- Madrid
- Stockholm
- Zurich
- London
North America (12)
- Montréal
- Toronto
- Querétaro
- Northern Virginia
- Columbus
- Iowa
- Dallas
- Las Vegas
- Los Angeles
- South Carolina
- Salt Lake City
- Oregon
South America (2)
- São Paulo
- Santiago
Asia (10)
- Hong Kong
- Delhi
- Mumbai
- Jakarta
- Osaka
- Tokyo
- Singapore
- Seoul
- Taiwan
- Bangkok
Oceania (2)
- Melbourne
- Sydney
Middle East (3)
- Tel Aviv
- Doha
- Dammam
Africa (1)
- Johannesburg
Tags
Equivalent services on other platforms
Isolated environments for executing untrusted or AI-generated code in Linux containers or isolated Workers, controlled through the Sandbox SDK's TypeScript API for commands, files and exposed services. GA on the Workers Paid plan since April 2026.
Isolated Firecracker microVMs for running untrusted or agent-generated code, with Linux images, managed or custom OCI images, snapshots, persistent sandboxes and drives; JS/Python SDKs and CLI; billed on active CPU and provisioned memory