Tencent Cloud Open Sources Cube Sandbox for Large Scale AI Agent Deployment

Tencent Cloud has made its Cube Sandbox fully open source under the Apache 2.0 license, announced in a press release. The release includes the complete sandbox-as-a-service system, not just the SDK, and is designed for large scale AI agent deployment.

Cube Sandbox combines hardware-level isolation with a cold start time below 60 milliseconds. It supports both the OpenAI Python SDK and E2B SDK, allowing developers to migrate existing agent runtimes without code changes. Built on MicroVM architecture, each sandbox runs its own guest operating system kernel through KVM virtualization, ensuring isolation between instances.

The system introduces five technical features: hardware-level isolation, sub-60ms cold start, lightweight memory usage below 5 MB per instance, large scale concurrent scheduling exceeding 100,000 instances, and event-level snapshot rollback for state recovery. Tencent Cloud stated that the snapshot rollback functionality will be open sourced once finalized.

Cube Sandbox is available for use across research, development, and enterprise production environments. It supports deployments without Kubernetes or vendor lock-in, and enterprises can maintain private installations for compliance. Tencent Cloud plans to integrate Cube with its TACO AI Acceleration Engine and FlexKV cache system to create a full stack infrastructure for secure and efficient agent operation.