Overview
ZhiGuo AgentArts is an enterprise-level agent platform launched by Huawei Cloud at the INSPIRE Innovators Conference on June 5, 2026, positioned as a 'Harness Engineering' implementation solution for the Agent era. Different from general LLM platforms or open-source Agent frameworks, it packages the entire set of engineering problems that need to be solved when running Agents in production environments: long-term task stability, enterprise-level security, deep industry knowledge, and full-link observability.
It also open-sources the openJiuwen version concurrently, with over 90% homology with the enterprise version, allowing enterprises to start with the open-source version and smoothly upgrade to the enterprise version for security compliance and full hosting capabilities. The concurrently released 'ZhiGuo Yuan' is a supporting Agentic entry point, aggregating full-stack Agentic cloud services + agents + hundreds of models and thousands of large models, with capabilities skillified and CLI-ified.
Key Features
- Production-Level Long-Term Tasks: Relying on Huawei Cloud's underlying AMS Agentic memory storage + AgentSphere sandbox, supports day-level long-term task continuous execution without dropping chains
- Enterprise-Level Security: Hardware encryption + data capsules + confidential inference gateway, combined with Huawei Cloud's 1037-day zero major incident operation experience
- Deep Industry Knowledge: Leveraging ModelArts Next MaaS model routing, can access 15+ SOTA models and automatically schedule by scenario (cost/effectiveness/balance three strategies)
- Full-Link Observability: Every step of the Agent has complete logs, progress tracking, and anomaly alerts, facilitating troubleshooting and auditing in production environments
- openJiuwen Open Source Version: Over 90% homology with the enterprise version, developers can first verify on the open-source version, then switch to the enterprise version for security/compliance/SLA guarantees
- ZhiGuo Yuan Agentic Entry Point: A unified entry point that is skillified and CLI-ified, aggregating cloud services + agents + large models, allowing developers to call them in one stop
Use Cases
- Enterprise internal business process automation (approvals, data processing, report generation)
- Agent deployment in highly regulated industries such as finance/healthcare/government
- Day-level long-term tasks (regular data collection, long-running monitoring, complex research)
- Cost-sensitive scenarios with mixed multi-model scheduling (automatically selecting the cheapest model by task)
- Customers who need both 'domestication' and 'Agent capabilities'
Pros
- Systematizes the process of 'running Agents in production environments', not just delivering a chat interface
- Underlying reliance on Huawei Cloud's complete infrastructure (100,000-card clusters, PB-level memory storage, 5 million tokens/second throughput per thousand cards)
- Simultaneously open-sources openJiuwen, zero cost for developers to start
- MaaS model routing makes mixed multi-model usage simple (average cost reduction of 20%)
- Clear domestication compliance attributes, friendly to central state-owned enterprises/government/financial customers
Pricing
The enterprise version is billed according to Huawei Cloud's standard methods (resource packages/pay-as-you-go), with costs including computing power + Agent platform + security compliance + technical support. The open-source version openJiuwen is completely free and can be self-deployed.
Summary
ZhiGuo AgentArts mainly targets two types of customers: enterprises that need to actually run Agents in production environments and have requirements for security compliance/SLA; and central state-owned enterprises/government/financial industries with clear demands for 'domestication'. If you are an individual developer conducting experiments, starting with the open-source version openJiuwen is sufficient; if you heavily rely on overseas model ecosystems like Anthropic/OpenAI, you may still need to evaluate migration costs.