- Documentation
Volcano Engine Function as a Service
User guide
Cloud sandbox
Sandbox images
Cloud sandbox
Sandbox images
Sandbox images
A sandbox image refers to the container image used to start a sandbox instance. You can select either custom images or ready-to-use public images.
Volcano Engine Function as a Service (veFaaS) provides multiple public images for different sandbox applications. All public images have been warmed up to ensure rapid image loading when creating sandbox applications, giving an out-of-the-box experience.
Currently available public image types include:
- All-in-One: An all-in-one runtime environment image that integrates various tools and enables efficient execution of collaborative tasks by AI agents across multiple scenarios.
- Code: A dedicated image pre-installed with mainstream compilers and code editors. It provides a secure, isolated environment for code compilation, running, and testing.
- Browser: An image featuring a headless browser engine and control APIs. It is suitable for browser-related tasks, such as automated web testing and data scraping.
- Computer: A standard Linux desktop/terminal image. It provides a complete Linux operating system environment and supports standard Shell commands and desktop interactions.
- Experimental: An experimental sandbox environment for integrating and testing cutting-edge or community-driven agent frameworks and runtimes (such as OpenCode and OpenClaw). It supports rapid validation of new architectures, new interaction modes, and capability orchestration mechanisms. Note that it may contain functional changes or unstable features.
- R2E-Gym-Subset: A subset of the R2E (Request-to-Execution) reinforcement learning environment. It provides the core R2E task environment for lightweight evaluation of agent execution and feedback capabilities in specific scenarios.
- WebArena: A web interaction evaluation environment for real-world websites and task scenarios built on the WebArena benchmark. It provides a full GUI browser interface for assessing agent capabilities in complex web navigation, information retrieval, and multi-step GUI operation tasks.
- SWE-bench: [Beta - apply for trial] A standardized testing environment designed to evaluate the ability of AI models to fix code defects in software engineering.
Custom images can be preheated to ensure sandbox applications start up within seconds.
tip
Image warm-up is an optimization technique in the container field that involves pre-pulling and caching required container images onto target nodes (such as worker nodes in a Kubernetes cluster and edge nodes) before application deployment or container startup. This avoids pulling images from remote repositories during deployment, thus reducing container startup time, reducing repository and network load, and enhancing deployment efficiency.
To create a custom image, follow the steps below:
- In the top navigation bar, select the target region.
- In the left navigation bar, click Cloud sandbox > Sandbox image.
- On the Sandbox image page, select the Custom image tab and click Image preheating.
- On the Image preheating page, specify which images to warm up. You can warm up multiple images at once.
- Click Start preheating to submit an image preheating task.
Description
If the task fails to be submitted, handle the error based on the error message. If the issue persists, submit a ticket to obtain technical support. - On the Custom image tab, you can view the status of warm-up tasks. The status Preheated indicates the image has been downloaded successfully.
Last updated: 2026.08.19 14:06:07