Tencent Cloud Cube Sandbox Goes Fully Open-Source, with Five Major Breakthroughs Enabling Large-Scale Agent Deployment

Tencent Cloud released Cube Sandbox as open source under the Apache 2.0 license on April 21, 2026, expanding beyond an SDK to include a sandbox-as-a-service stack. The update centers on providing an architecture for deploying AI Agents with hardware isolation and performance-focused startup and scheduling.

In its description of the project, Tencent Cloud said Cube combines hardware-level isolation with sub-60ms cold start and supports OpenAI Python SDK and E2B SDK usage. The company also tied the release to an internal AI Agent infrastructure strategy and said Cube is available for one-click deployment.

Tencent Cloud said Cube is built on a microVM architecture and runs a dedicated Guest OS kernel via KVM hardware virtualization, with no shared kernel. It reported cold start performance under 60ms using resource pool pre-provisioning, snapshot cloning, EPT lazy load, and lock optimization, and said per-instance memory overhead is under 5MB. The company described distributed scheduling and bin-packing for burst scheduling of 100K+ instances and an event-level snapshot rollback capability that supports checkpoint saving, arbitrary state rollback, and rapid forking.

The company said Cube Sandbox includes a one-click script for setup, with integration through MCP, API, SDK, or CLI requiring zero rewrites to Agent code. It also stated the project released its full codebase, deployment scripts, documentation, and examples covering Shell execution, file operations, browser automation, and RL training. Dowson Tong, Senior Executive Vice President of Tencent and CEO of the Cloud and Smart Industries Group, said, “A breach in one sandbox leaves the rest untouched.” Tencent Cloud also said it will pair Cube with TACO AI Acceleration Engine and FlexKV cache system in a combined infrastructure for secure sandbox runtime, inference acceleration, and cache optimization.