ME News reports that on April 21 (UTC+8), according to monitoring by Beating, Tencent Cloud has open-sourced the AI Agent sandbox Cube Sandbox, written in Rust under the Apache 2.0 license. A sandbox is an isolated environment for running agent-generated code, preventing accidental file deletion or unauthorized host access. Products such as OpenAI Agents SDK, Manus, Perplexity, and Hugging Face currently use similar architectures, with E2B serving as the de facto interface standard. Cube is compatible with the E2B interface—no changes to business code are required; simply modifying one environment variable enables switching from E2B’s hosted service to a self-deployed Cube instance. Tencent Cloud has published two sets of performance metrics. Single-concurrency cold start is under 60ms; at 50 concurrent instances, average latency is 67ms, P95 is 90ms, and P99 is 137ms. Resident memory per instance is under 5MB (measured with sandbox configurations up to 32GB), enabling over 2,000 sandboxes to run simultaneously on a single 96-core server. In comparable scenarios, Docker containers require approximately 200ms to start and share the host kernel; traditional virtual machines take seconds to boot and consume at least 20MB of memory per instance. Cube achieves this by assigning each Agent its own independent guest OS kernel with hardware-level isolation, while reducing startup time to under 100ms. Speed improvements are enabled through resource pool pre-allocation, snapshot cloning, and底层 lock optimizations; memory reduction is achieved via Rust reimplementation, Copy-on-Write (CoW) memory reuse, and reflink-based disk sharing. The project also includes CubeVS, which uses eBPF for network isolation between sandboxes. Two large-scale validation cases have been provided. Originally running within Tencent Cloud’s Serverless infrastructure, Cube has handled tens of billions of requests. After migrating the Yuanbao AI coding scenario to Cube, resource consumption dropped by 95.8%. Among external customers, MiniMax leveraged Cube to schedule hundreds of thousands of sandbox instances within minutes for Agentic RL training. The next planned step is to open-source event-level snapshot rollback functionality, enabling state rollback within hundreds of milliseconds. (Source: BlockBeats)
Tencent Cloud open-sources CubeSandbox, supports E2B, and runs 2,000 sandboxes on a single machine
KuCoinFlashShare
Summary
On April 21 (UTC+8), Tencent Cloud announced the open-sourcing of CubeSandbox, an AI agent sandbox written in Rust and licensed under Apache 2.0. The tool supports E2B compatibility and enables 2,000 sandboxes to run on a single 96-core server. On-chain news platforms may benefit from its low-latency execution and secure environment. A case study demonstrated a 95.8% reduction in resource usage for the Yuanbao AI programming scenario. Inflation data tracking and other compute-intensive tasks could achieve performance improvements with CubeSandbox deployment.
Source:Show original
Disclaimer: The information on this page may have been obtained from third parties and does not necessarily reflect the views or opinions of KuCoin. This content is provided for general informational purposes only, without any representation or warranty of any kind, nor shall it be construed as financial or investment advice. KuCoin shall not be liable for any errors or omissions, or for any outcomes resulting from the use of this information.
Investments in digital assets can be risky. Please carefully evaluate the risks of a product and your risk tolerance based on your own financial circumstances. For more information, please refer to our Terms of Use and Risk Disclosure.