10月20日，国际企业IT管理解决方案企业ManageEngine卓豪面向中国市场推出30余款Zia智雅智能体，并计划到2026年底扩展至100余款，旨在强化IT管理全栈能力、常规任务自动化处理能力，并实现更高效、更具自主性的业务执行。这些智能体由自研Zia智雅大模型驱动，不产生额外Token成本，直接集成在客户已使用的产品内，可直接复用各产品已有的业务运行上下文，无需额外采购AI平台或智能体工作流工具，也无需复杂配置。
On October 20, international enterprise IT management solution company ManageEngine launched over 30 Zia intelligent agents for the Chinese market, planning to expand to over 100 by the end of 2026, aiming to strengthen full-stack IT management capabilities, routine task automation, and more efficient, autonomous business execution. These agents are driven by the self-developed Zia large model, generating no additional Token costs, directly integrated into products customers already use, and can directly reuse each product's existing business operation context without additional AI platform or agent workflow tool procurement or complex configuration.
该系列智能体将于今年10月28日在深圳举办的ManageEngine卓豪用户大会上正式亮相。这是ManageEngine卓豪继今年5月在全球推出智能体后，面向中国市场的深度本土化落地。依托其自研的Zia智雅大模型及位于北京、上海的本地数据中心，ManageEngine卓豪正在加速中国企业IT运维的智能化升级。其产品全面适配国内企业IT基础设施需求，支持人大金仓数据库、麒麟和欧拉操作系统、Arm架构、东方通中间件，并兼容DeepSeek等主流模型。
This series of agents will officially debut at the ManageEngine User Conference in Shenzhen on October 28. This is ManageEngine's deep localization landing for the Chinese market after globally launching agents in May. Relying on its self-developed Zia model and local data centers in Beijing and Shanghai, ManageEngine is accelerating Chinese enterprises' IT operations intelligent upgrade. Its products fully adapt to domestic enterprise IT infrastructure needs, supporting Kingbase database, Kylin and openEuler operating systems, Arm architecture, TongTech middleware, and compatible with mainstream models including DeepSeek.
埃森哲中国调研显示，63%的中国企业已部署AI助手或AI驱动的知识管理系统，而一年前该比例仅为28%。Gartner对IT负责人的调研显示，75%的IT管理者正在试点、部署或已经落地各类AI智能体。然而，规模化落地智能体AI也带来了成本和数据隐私方面的全新挑战。许多AI解决方案按Token、单次操作或使用量计费，随着采用规模扩大，成本不断攀升。此外，依赖外部第三方模型可能会带来企业敏感数据泄露风险，而运行私有大模型则需要投入GPU基础设施、运维资源和专业技术人才。
Accenture China research shows 63% of Chinese enterprises have deployed AI assistants or AI-driven knowledge management systems, versus only 28% a year ago. Gartner's IT leader survey shows 75% of IT managers are piloting, deploying, or have landed various AI agents. However, scaled agent AI deployment brings new cost and data privacy challenges. Many AI solutions charge by Token, single operation, or usage volume, with costs climbing as adoption scales. Additionally, relying on external third-party models may bring sensitive enterprise data leakage risks, while running private large models requires GPU infrastructure, operations resources, and professional technical talent.
据ManageEngine卓豪全球副总裁Mathivanan分享，能实现零Token成本智能体AI，主要依托三大核心优势，30年的工程技术积淀，经过验证的自研AI模型，以及自有运营的基础设施。Zia智雅大模型为卓豪公司全系列产品中的AI功能提供支撑，服务超1.5亿用户，支持120余款产品。他说，ManageEngine卓豪希望能够让客户的数据留在内网不会泄漏出去。当客户接入DeepSeek这类公共大模型时，其数据仍能保持隐私性，同时ManageEngine卓豪配备的报警系统也会进一步保障数据隐私安全。
ManageEngine Global Vice President Mathivanan shared that achieving zero-Token-cost agent AI relies on three core advantages—30 years of engineering accumulation, validated self-developed AI models, and self-operated infrastructure. The Zia large model supports AI functions across ManageEngine's full product line, serving over 150 million users across 120+ products. He said ManageEngine hopes to keep customer data on the intranet without leakage. When customers connect to public large models like DeepSeek, their data remains private, and ManageEngine's alarm system further safeguards data privacy.
ManageEngine卓豪此次推出的30余款Zia智雅智能体，深度嵌入其五大核心解决方案领域，全栈可观测性、统一服务管理、统一端点管理与安全、身份与访问管理、安全信息与事件管理，支持云、本地部署与托管服务提供商环境，实现检测、调查、关联分析、问题处置、归档的端到端完整工作链路。本次推出的智能体包含根因分析智能体、应用性能监控助手、一线服务台专家、知识库文章生成助手、设备故障排查智能体和资源纳管等，能够将平均故障解决时间缩短40%至60%，可自主处理40%至60%的一线服务台工单，将原本耗费数小时的排查工作压缩至分钟级。
ManageEngine's 30+ Zia agents deeply embed into its five core solution areas—full-stack observability, unified service management, unified endpoint management and security, identity and access management, and security information and event management—supporting cloud, on-premises, and managed service provider environments, achieving end-to-end workflows from detection and investigation to correlation analysis, problem resolution, and archiving. The agents include root cause analysis agents, application performance monitoring assistants, frontline service desk experts, knowledge base article generation assistants, device troubleshooting agents, and resource management, capable of shortening mean time to resolution by 40% to 60%, autonomously handling 40% to 60% of frontline service desk tickets, compressing hours-long troubleshooting to minutes.
ManageEngine卓豪中国首席运营官李飞说，据测算，大型企业每年在智能体AI推理上的成本可达20万至40万元人民币。Zia智雅大模型驱动的智能体不再收取额外的Token费用，希望以此帮助中国企业规模化落地AI，降低运营成本。Mathivanan谈道，ManageEngine卓豪的差异化优势包括实现智能体产品覆盖全业务场景、可实现内部操作的联动，以及通过分析内部系统数据进行风险预测。李飞称，ManageEngine卓豪开发的智能体在安全审计、回滚机制、安全护栏及权限设计等方面均有充分保障。
ManageEngine China COO Li Fei said calculations show large enterprises' annual agent AI inference costs can reach 200,000 to 400,000 yuan. Zia model-driven agents no longer charge additional Token fees, hoping to help Chinese enterprises scale AI deployment and reduce operating costs. Mathivanan said ManageEngine's differentiation includes agent products covering all business scenarios, enabling internal operation linkage, and risk prediction through analyzing internal system data. Li Fei said ManageEngine-developed agents have full guarantees in security auditing, rollback mechanisms, safety guardrails, and permission design.
这条新闻和前面讨论的具身智能、机器人产业看起来离得比较远，但它反映的是同一个趋势的不同侧面。前面讨论过FDE、数据采集员、机器人训练场、开源体系，这些都是具身智能落地的配套环节。ManageEngine卓豪做的是企业IT运维的智能体，和机器人没有直接关系，但它处理的问题和具身智能落地面临的问题有相似之处。智能体要真正在企业里干活，需要解决成本、数据隐私、安全审计、权限设计这些工程问题，而不是只靠模型能力。这和前面讨论的机器人要真正在工厂里干活，需要解决感知、理解、决策、远程接管、数据闭环这些工程问题，是同一个逻辑。AI从演示走向生产环境，决定成败的往往不是模型本身，而是这些配套的工程体系。
This news seems far from the previously discussed embodied intelligence and robot industry, but it reflects different sides of the same trend. Earlier we discussed FDE, data collectors, robot training grounds, and open-source systems—all supporting links for embodied intelligence deployment. ManageEngine does enterprise IT operations agents, not directly related to robots, but the problems it handles resemble those embodied intelligence deployment faces. For agents to truly work in enterprises, cost, data privacy, security auditing, and permission design must be solved—not just model capability. This follows the same logic as the previously discussed robots needing to solve perception, understanding, decision-making, remote takeover, and data loops to truly work in factories. When AI moves from demonstration to production environments, what determines success is often not the model itself but these supporting engineering systems.
