openJiuwen Releases Enterprise-Grade AgentOS as Open Source to Scale Agent Adoption

The project aims to connect upper-layer agent applications with underlying models and compute resources through a plugin-based architecture and an open ecosystem. Its developers position it as a system that addresses problems such as long and complex task chains, inefficient coordination among multiple agents, difficulty retaining business experience, rising compute costs and concerns over data permissions, security isolation and fault recovery.

In multi-agent collaboration, AgentOS uses swarm collaboration and workflow orchestration to support task decomposition, role assignment and parallel execution. It can organize stages such as research, analysis and writing, while allowing humans to participate in decisions at key points so that agent-team collaboration remains controllable.

For self-evolution, the platform relies on execution traces, user feedback and memory mechanisms. It allows agents to improve their skills and collaboration methods and turn effective experience into reusable capabilities, reducing repeated trial and error as business requirements change.

The enterprise runtime foundation focuses on compute affinity and stability. Through hardware-software co-design, resource scheduling and inference optimization, AgentOS seeks to improve compute utilization. It also provides multi-tenant isolation, secure sandboxes, permission management and safety guardrails to constrain data access and tool calls, and uses fault recovery mechanisms to support continuous, reliable task execution.

AgentOS is plugin-based and has an open northbound ecosystem. Model invocation, tool invocation and memory management can be loaded as plugins and extended as needed. In the northbound ecosystem, five types of assets, including skills, connectors, plugins, experts and expert teams, can be combined, published, acquired and shared through Agentic Hub. Partners can reuse existing capabilities or turn industry methods, system connections and team collaboration experience into their own assets, reducing duplicate development and speeding the rollout of industry agents and all-in-one solutions.

At Huawei Connect 2026, Huawei released an open-source Agent acceleration platform and solutions based on Jiuwen AgentOS, together with a series of hardware products, to work with partners on industry agent all-in-one machines. Jiuwen AgentOS provides a high-reliability and high-security runtime foundation for the acceleration platform. It includes a unified Agent gateway, a distributed runtime foundation and a built-in WorkSwarm unified workbench for office and programming tasks, reducing the need for enterprises to build and adapt components themselves. A built-in SystemAgent supports one-click configuration and management of the all-in-one machine. According to information released at the conference, the platform can enable hour-level end-to-end private deployment for enterprises.

For industry customization, partners can use open plugins and interfaces to connect existing business systems, local knowledge bases and industry models, and to bring their own expertise into AgentOS. For example, for internal information retrieval, analysis and report generation, partners can configure collaborative agents to make them closer to actual business workflows. As demand grows, they can add more skills and applications.

Jiuwen AgentOS is intended to give partners a reusable technology foundation and enterprises an agent application environment that is easier to deploy, manage and expand. The project is available at https://atomgit.com/openJiuwen/agentos-enterprise and https://github.com/openJiuwen-ai/agentos-enterprise, QbitAI reported.