At the 2025 World Artificial Intelligence Conference (WAIC), PPIO, an independent distributed cloud computing service provider, made a groundbreaking announcement with the launch of China's first Agentic AI infrastructure service platform. This marks a pivotal moment in the development of intelligent agent applications, ushering in a new era of growth and innovation.
PPIO's CEO and co-founder, Yao Xin, unveiled the AI intelligent agent platform, which is divided into two major product lines: General Edition and Enterprise Edition. The General Edition, powered by a distributed GPU cloud base, offers a cost-effective, efficient, and high-performance development environment for individual developers and small to medium-sized enterprises. On the other hand, the Enterprise Edition caters to medium to large enterprises by leveraging five key differentiating advantages: long-term memory, controllable processes, full-domain operations, business linkage, and multimodal collaboration, creating a closed-loop system from intelligent decision-making to execution.
PPIO's Agent Sandbox, a cloud-based execution environment designed for intelligent agents, is built on Firecracker MicroVM and boasts three core features: strong security isolation to prevent data leaks and resource conflicts between different agent sandbox environments; sub-200ms millisecond-level ultra-fast startup to enhance development fluidity; and high concurrency creation capability supporting thousands of sandbox instances to meet high business concurrency demands.
The sandbox supports dynamic invocation of various tools such as Browser use, Computer use, MCP, RAG, Search, and has integrated with renowned open-source projects like Camel AI, OpenManus, and Dify, endowing agents with secure, efficient, and agile execution capabilities.
In terms of agent decision-making requirements, PPIO's model service supports rapid access to hundreds of mainstream open-source and customized AI models. It has launched domestic mainstream inference models like DeepSeek R1, Qwen3, and MiniMax M1 for planning capabilities; upgraded DeepSeek's context window to 160K to meet multi-round, ultra-long conversation needs for memory requirements; and supported code models like Kimi-K2 and Qwen3Coder for execution capabilities, enabling powerful autonomous programming and tool invocation.
The platform also offers multimodal model support for text, voice, video, and images, and has deeply optimized its own inference acceleration engine with core technologies like PD separation, FlashMLA, and DeepEP to achieve a dynamic balance between inference efficiency and resource usage.
PPIO's Enterprise Edition has been validated in real-world scenarios, such as creating a comprehensive smart education solution for an international school. This solution covers the entire spectrum of management, teaching, learning, and research, spanning the entire process from pre-class preparation, in-class assignments, learning analytics, psychological monitoring, to home-school interaction, adapting to various educational stages both inside and outside the school.
Yao Xin stated, "From AI Infra to Agentic AI Infra, PPIO's AI intelligent agent platform not only provides comprehensive support from underlying computing power to upper-layer applications for agent developers and enterprises, reducing technical barriers and development costs, but also lays a solid foundation for the large-scale deployment of agents through a secure and efficient sandbox environment, diverse model resources, and cost-effective computing services, propelling agents from technological exploration to a new stage of value creation."
With the launch of this platform, intelligent agent application development will gain stronger infrastructure support, which is expected to accelerate the large-scale application and commercial value realization of agent technology across various industries.