Docs

Volcano Engine Function as a Service

User guide

Selection guide

veFaaS is an event-driven, fully managed serverless computing platform. Built on a unified computing foundation, veFaaS offers three core product forms: applications, functions, and cloud sandboxes, to address the needs of various scenarios, from lightweight tasks to complex AI applications. Understanding their differences and connections is essential to leveraging the serverless architecture efficiently and accelerating business innovation.

Application: A one-stop, scenario-oriented serverless solution

A veFaaS application is a highly encapsulated, scenario-oriented serverless form. It packages underlying functions, triggers, dependent environments, and infrastructure configurations (such as GPU and network resources) into a ready-to-use solution. Through simple console operations or API calls, developers can deploy a fully functional application with a single click.

You can consider an application as a best-practice template and a one-stop deployment unit provided by veFaaS. veFaaS applications significantly lower the barrier to using serverless. They are especially suitable for scenarios focused on rapid implementation with minimal attention to underlying details.

Core features

Scenario-based encapsulation: Each application is designed for a specific scenario, such as an AI agent, Web service, or AI painting (Stable Diffusion).

Quick deployment: Pre-built application templates allow you to deploy services in minutes without configuring resources from scratch.

‌Zero O&M: veFaaS applications inherit the core advantages of serverless to enable automated scaling, load balancing, and resource management.

Built-in integration: Applications provide preset integration with Volcano Engine cloud services, such as TOS and API Gateway. This integration support offers high compatibility within the ecosystem and improves development efficiency.

Application scenarios

veFaaS applications are particularly applicable to the following scenarios:

Rapid POC for cloud-native AI applications: This is the core scenario for the application form. veFaaS provides a wide range of AI application templates. Examples include:

AI Agent and MCP Server: Provide a serverless hosting environment for large language model (LLM) applications, such as Browser Use Agent and Computer Use Agent. This enables agents to perform complex tasks involving interactions with browsers and desktop environments.

GPU inference and training: Enable quick deployment of GPU-accelerated AIGC applications, such as Stable Diffusion (an AI painting application) and ComfyUI. You can use expensive GPU resources on demand.

Coze application hosting: Seamlessly deploy AI Bots created on the Coze platform and provide them as independent, external services.

Standard web services and APIs: Quickly build web applications or RESTful APIs that comply with standard frameworks (such as Express). They are suitable for small and medium-sized projects, prototyping, and internal tools.

Lightweight data and media processing: For example, use FFmpeg application templates for media processing tasks, such as video frame extraction and format conversion.

Function: An event-driven computing unit

A veFaaS function is the most fundamental and core form of serverless computing, commonly known in the industry as FaaS (Function as a Service). It allows developers to focus solely on the core business logic code (a function). Unlike highly encapsulated applications, functions offer maximum flexibility and customization. Developers can start from scratch to freely combine triggers and write code logic of any complexity.

Core features

Event-driven: Functions are triggered by explicit event sources, such as an API gateway receiving an HTTP request, a message queue (Kafka/RocketMQ) receiving a new message, a TOS bucket receiving a new file upload, or a timer expiring.

Stateless and short-lived execution: Function instances are stateless by default and are typically designed to complete a task within a limited time (from a few seconds to a few minutes). State persistence relies on external storage, such as databases or caches.

On-demand execution and automatic scaling: The platform automatically creates or destroys function instances based on real-time request volume, achieving millisecond-level scaling.

Fine-grained billing: Costs are based solely on the actual number of function executions and the computing resources consumed (CPU × duration, memory × duration). You pay only for what you use.

Application scenarios

veFaaS functions are efficient tools for building modern serverless backends and are suitable for almost all backend scenarios that do not require long-running processes or persistent connections:

Web APIs and microservices: As a core component of the backend microservices architecture, each function is an independent API endpoint or business capability. Exposing services externally through an API gateway is an ideal choice for building highly resilient and cost-effective web services.

Data processing and ETL: Perform real-time extract, transform, and load (ETL) operations on data from various sources, such as Kafka, RocketMQ, and TOS. For example, uploading a picture to TOS automatically triggers a function to compress it, add a watermark, and store it in another location.

Asynchronous task processing: Execute long-running backend tasks, such as sending emails, generating reports, and performing data analysis. After the client submits a task, a response is returned immediately while the function executes asynchronously in the backend. Upon completion, a callback can be made or a message can be sent as a notification.

Scheduled tasks and automated O&M: Use a timer trigger to periodically perform tasks such as database backups, report generation, system health checking, and cleanup of expired data.

Internet of Things (IoT) backend: Process numerous data streams from IoT devices for real-time analysis, alerting, and storage.

Cloud sandbox: A secure, isolated execution environment for AI agent applications

A veFaaS cloud sandbox is a serverless form that provides a secure, isolated execution environment. It allows you to start a temporary, isolated container environment (sandbox instance) dynamically and on demand to execute untrusted code or perform tasks with complex environment dependencies. Upon task completion, this environment is destroyed immediately to ensure security and cost-effectiveness.

Unlike applications and functions, which are primarily service-oriented, cloud sandboxes focus on isolation, security, and dynamism. A cloud sandbox is more like a serverless version of a temporary virtual machine or container, designed specifically for scenarios requiring dynamic and secure execution environments.

Core features

Strong isolation and security: Each sandbox instance runs in an independent virtualized environment, which is strictly isolated from the host and other instances. This isolation prevents malicious code escapes and lateral attacks.

Dynamic creation and destruction: Sandbox instances can be created on demand via API calls to execute tasks. These instances are automatically destroyed after tasks are complete or time out, with resources released immediately.

Environment customization: You can create sandbox runtime environments using custom container images. This allows you to preinstall any tools, libraries, and dependencies to meet complex task requirements.

API-driven: The entire lifecycle (creation, command execution, and destruction) is fully controlled by APIs, which enables easy integration with existing business systems.

Application scenarios

A cloud sandbox is an ideal solution to needs such as "providing users with a programmable and executable environment":

Tool execution environments for AI agents: Provide a secure execution environment for AI agents. When an agent needs to execute code, access the file system, or interact with external tools, it can do so in the sandbox. Even if risky operations are performed, the core system remains unaffected.

Reinforcement learning and evaluation: Address scenarios requiring rapid and batch creation of agent runtime environments and evaluation environments (for example, Swe-bench evaluation).

Code sandbox for chatbots and online programming: Provide users with an online code editor and runtime environment. User-submitted code is executed in an isolated sandbox, and the execution results are returned to the frontend. This ensures platform security.

Comparison

The following table provides a multidimensional comparison of the three to help you understand their differences:

Dimension	Application	Function	Cloud sandbox
Core value	Scenario-based solution and quick deployment	Fundamental unit of event-driven computing that offers maximum scalability	Securely isolated dynamic execution environment
Application scenarios	AI agents, GPU-accelerated inference, standard web services, and template-based scenarios	Web APIs, microservices, data processing (ETL), scheduled tasks, and asynchronous tasks	Tool execution for AI agents, online programming, secure code execution, and CI/CD
Trigger/Access method	Usually accessed directly via a public URL (built-in API Gateway)	Driven by events (for example, API Gateway, MQ, TOS, and Timer events)	Active OpenAPI/SDK invocations to create and control instances
Scaling	Fully automated scaling based on traffic patterns	Fully automated scaling based on event frequency or request volume (instance reservation can be configured)	On-demand, manual, or programmatic creation and destruction of instances
State management	Stateless by default. Some applications (such as MCP Server) provide session persistence mechanisms	Strictly stateless (relying on external storage, such as Redis and other databases)	Stateful. Sandbox instances can maintain state during their lifecycle (such as file system and memory)
Cost and billing	Billed based on consumed computing resources (vCPU/memory/GPU) and invocation count	Billed based on function execution count and computing resources (Unit cost of vCPU/memory/GPU * seconds)	Billed based on the runtime and specifications of sandbox instances
Customization level	Low (restricted by template capabilities)	Extremely high (full customization support)	High (implemented via custom images and API control)

Last updated: 2026.08.19 14:06:08