Shizai Intelligence Launches Enterprise-Grade AI Agent Platform Soiawork

Recently, Shizai Intelligence officially launched its enterprise-grade AI agent platform Soiawork (Shuya Work). The platform is aimed at building enterprise AI organizations, and attempts to let AI reach real business operations through connectors and computer operation capabilities, to enable continuous reuse of enterprise knowledge and work methods through knowledge bases, Skills and Workflows, and to allow team members and AI to carry out tasks under a shared context through shared workspaces. On this basis, enterprises can also use permissions, approvals, auditing, and usage and cost management to give AI use boundaries and keep records of investment, providing a basis for evaluating business value.

The release of Soiawork responds to several practical problems in the process of enterprise AI implementation: if a system has no interface, does AI have no entry point; how can an AI that can do things act according to the enterprise's rules; and for the same task, how can team members and AI quickly align. Shizai Intelligence stated that Soiawork hopes to start from a single task, let AI enter business operations, and turn effective experience into organizational capability.

AI can understand tasks, but it still needs to obtain relevant information and call tools before it can move work forward. Enterprises' daily business is distributed across office platforms, CRM, ERP and various industry software. Some of these provide open interfaces, while others still require employees to open interfaces, query records and fill in information.

In response to this situation, Soiawork uses two types of capabilities to connect to the business frontline. For software that provides open interfaces such as APIs and MCP, Soia can use connectors to invoke supported operations such as query, create and update within the authorized scope. For systems that lack usable interfaces and require interface operations, Soia can invoke the Worker installed on the execution device, combining RPA and Computer Use Agent capabilities to complete tasks within the scope of adaptation. According to Shizai Intelligence, this capability achieved a 90.2% task success rate on the OSWorld benchmark and once topped the global overall ranking.

In addition, users can create custom connectors by conversing with Soia, and the enterprise edition also provides exclusive software self-learning capabilities to complete deep adaptation of complex systems. The two types of capabilities, connectors and Worker, complement each other, giving more existing business systems the opportunity to participate in AI tasks. At present, Soiawork's connectors mainly target business scenarios such as manufacturing, domestic e-commerce and cross-border e-commerce, covering MES, ERP, CRM, SRM, office collaboration, enterprise information query and related business tools. Enterprises can look up existing connection methods in the connector directory, or configure custom connectors according to their own systems and task requirements. The specific scope of operations after connection is jointly determined by the capabilities supported by the connector, the system adaptation situation and valid authorization.

Taking a recruitment task as an example, after a recruiter receives a job JD, they need to sort out the talent profile, review candidate resumes, screen candidates who meet the requirements, and then prepare personalized communication content. These steps require both understanding business requirements and actually operating the recruitment platform. Taking the interface operation scenario of BOSS Zhipin as an example, Soiawork can combine with Worker to assist in completing talent profile analysis, candidate screening, resume review and personalized greetings, and organize communication records by position. The recruiter can stop execution and retains the final review authority. What this case demonstrates is how AI connects understanding tasks, obtaining information, operating software and organizing results.

Knowledge bases, Skills and Workflows keep enterprise experience

In an enterprise, how a piece of work is done often involves three types of accumulation: what business materials can be referenced, what proven methods exist, and which steps and rules must be followed. Through knowledge bases, Skills and Workflows, Soiawork links these accumulations with actual tasks.

In terms of knowledge bases, product manuals, rules and regulations, historical documents and frequently asked questions can be organized into searchable knowledge content. Employees ask questions through conversation and obtain relevant information and source references. The file space supports uploading files and also supports importing documents from DingTalk and Feishu; materials generated during tasks can be retained for subsequent use. Creating a knowledge base can also start from a conversation, and Soiawork will assist in determining the scope of materials and the organization method based on existing materials and the purpose explained by the user. After materials are updated, they can participate in subsequent retrieval after synchronization processing is completed.

Skills are used to leave the method of accomplishing something for the next time. For example, e-commerce operations need to organize business data every week, screen abnormal products, and then generate a to-do list. This work has fixed data standards, screening rules, processing steps and delivery formats, and these requirements can be consolidated into a Skill for invocation in subsequent tasks. A Skill can carry task steps, business rules, scripts, templates and reference materials. Enterprises can organize methods through natural language, or extract steps from effective tasks, save them after confirmation, and then continuously test, improve and share them.

Workflows are used to organize tasks that include multiple stages, judgment conditions and input/output requirements. For example, a business analysis task can sequentially obtain data, clean details, judge anomalies, generate results, and then output a to-do list. Different inputs enter corresponding paths according to set conditions, and abnormal situations are handled according to configuration. Soiawork supports generating or adjusting workflows through natural language, as well as testing, publishing and repeatedly running them. Material processing, data processing, knowledge retrieval, and supported connectors and computer operations can all become stages in the process.

Taking Amazon operations as an example, a team can organize its methods for checking listing quality into a Skill, specifying the check metrics, scoring criteria and report format, then combine them with already-connected data and applicable processes to generate check results and optimization suggestions. Verified methods can continue to be reused and continuously improved as the business changes.

Someone in the project group reports, "This feature isn't working, please take a look." Next, the team may need to supplement the issue information, look up product materials and historical records, register a ticket, and then have the relevant personnel follow up. When different members and AI participate in the same type of work, they need to understand a shared context and also need to use the corresponding materials, methods and tools.

Soiawork's workspace can organize this shared information and these resources in a centralized way. Teams can record business context, work requirements and execution boundaries in the space description, associate files, knowledge bases, Skills, Workflows, connectors, and shared devices with Workers installed as needed, and configure the corresponding usage permissions. For example, for handling project issues, a team can create a dedicated workspace, put in handling specifications, product documentation and historical materials, and associate the required skills, processes and system connectors. When members initiate related tasks in the space, Soia can carry out the work based on the existing context and authorized resources, reducing repeated explanations and resource configuration. The creation and configuration of a workspace can also be completed step by step through conversation.

As AI moves from individual experimentation to team application, enterprises also need to answer: where is AI used, how much does it cost, and are these investments worth continuing. Soiawork's enterprise management backend helps enterprises see usage clearly and manage members, resources and the execution process.

Enterprises can view the organization's credit balance, validity period and related transactions, and analyze usage and consumption by task, executing entity and billing item. Managers can thus understand which tasks AI resources are mainly used for, providing a reference for expense reconciliation, resource planning and input-output evaluation. Only by combining these records with business outcomes such as the enterprise's actual task completion, processing time and rework can they further judge which tasks are worth continued investment, which methods are suitable for promotion, and which stages still need improvement.

In terms of permissions, administrators can manage organization members, roles, departments and member groups, configure permissions according to responsibilities, and manage login and organizational access through supported enterprise identity access methods. Resources such as connectors, skills, automated tasks and shared devices can also have view, use or manage permissions configured as needed. Operations that require confirmation or approval can enter the corresponding stage according to configuration. In terms of process records, through organization task monitoring, administrators can query running status, view related execution events, result files and failure reasons, helping to locate problems, review results and improve processes. Operation auditing provides records for tracing related behavior.

Soia's name comes from thinking about AIOS (AI Operating System, an enterprise-level AI operating system), and also embodies Intellective Intelligence's thinking and accumulation on AI organizations over eight years of serving enterprises in AI and automation. Soiawork is launched by Intellective Intelligence and is an independent product for building enterprise AI organizations. It differs from Intellective Agent in positioning and capability system, and enterprises can choose according to actual needs. Intellective Intelligence was founded in 2018, has long been deeply engaged in the AI and automation field, and has served more than 6,000 enterprises cumulatively.

Intellective Intelligence stated that the so-called AI organization is one in which people and AI, around business tasks, share available knowledge, methods and tools, collaborate under clear permission and management mechanisms, and continuously evaluate the value of this work. This path can start from a specific task: choose a piece of work that is currently happening, clarify the goals, required resources and acceptance criteria; verify results and record consumption during actual execution, retain effective methods, and then gradually expand to more tasks and teams.

Whitelist source: