StaffDeck – An Open-Source Enterprise-Level Digital Employee Platform by Baichuan Intelligence

1. What is StaffDeck

StaffDeck is an open-source enterprise-level platform for the full-process construction and management of digital employees, jointly developed by Baichuan Intelligence, the Northeastern University-Baichuan Intelligence Data Intelligence Joint Laboratory, the THUNLP Lab at Tsinghua University, OpenBMB, and AI9Stars. This platform transcends the traditional AI assistant paradigm by treating Agents as formal employees with positions, employee IDs, SOPs, knowledge bases, and performance records. Through a state machine-driven process engine, structured knowledge base, and feedback iteration loop, it enables the evolution from a mere conversational tool to a capable, experience-accumulating digital workforce, offering enterprises a standardized and governable AI workforce management solution.

Image source: Official article
Image source: official article

Technical positioning and domain: StaffDeck is an enterprise-level AI Agent operations platform, focusing on the full lifecycle management of digital employees, including creation, configuration, deployment, monitoring, and continuous optimization. Unlike general-purpose RPA or chatbots, it emphasizes "employee-style governance," integrating Agents into the enterprise human resources management system and granting them formal identities and performance records, thereby enabling asset accumulation and compliance control at the organizational level.

Development background: This project was led by Baichuan Intelligence in collaboration with the Northeastern University Data Intelligence Joint Laboratory, the THUNLP Lab at Tsinghua University, the OpenBMB open-source community, and AI9Stars. Baichuan Intelligence has deep expertise in AI Agent and open-source large models, with its MiniCPM series providing the lightweight inference foundation for the platform. The OpenBMB community contributed mature toolchains and ecosystem resources, jointly advancing the engineering implementation of digital employees from concept to reality.

Core value: StaffDeck addresses key challenges in enterprise AI application deployment, such as Agent management difficulties, rigid workflows, and non-traceable knowledge. Through employee records, state machine SOPs, OKF structured knowledge base, and a full-chain feedback loop, the platform enables Agents to be evaluated, accumulate experience, and continuously improve, just like formal employees, significantly reducing the cost of deploying and maintaining AI labor for enterprises.

Technical features: It introduces a unique state machine-driven process skill, supporting the serial execution of multiple SOPs, interruption during execution, and reuse of context, balancing process determinism with interactive flexibility. It incorporates the OKF five-tier knowledge standard to ensure that every response can be traced back to specific document sections. A full-chain trace and negative feedback automatic attribution mechanism forms a rapid iteration loop of "today's mistake, tomorrow's correction," preventing Agents from repeating errors.

2. Key Features

• Digital Employee Marketplace: Establish complete employee profiles for each Agent, including job roles, skills, SOPs, work records, and performance data. Support unified management by department and role, enabling organizational-level Agent asset accumulation, facilitating human resource-level allocation and evaluation, and aligning digital employee management with real employee systems.

• One-Sentence SOP Generation: Business personnel can describe processes in natural language, and the system automatically distills them into visual state machine flowcharts. After confirmation, they are published as procedural skills for digital employees, significantly reducing the barrier for process configuration and allowing non-technical staff to quickly build complex business logic, directly converting experience into executable automated workflows.

• Multi-SOP Sequential Execution: Multiple processes can be executed sequentially within the same conversation, reusing context information already obtained without requiring users to re-enter data. This ensures a consistent experience across process tasks and improves processing efficiency in complex business scenarios, such as automatically linking approval and payment steps in a reimbursement process.

• Interrupt Handling and Automatic Recovery: When temporary issues arise during execution, the process is paused and the state is saved. After resolution, the Agent resumes execution from the interruption point. This mechanism grants digital employees flexibility similar to humans, preventing process resets due to unexpected interruptions, especially useful in scenarios requiring human intervention such as approvals or exception handling.

• OKF Structured Knowledge Base: Introduce an open knowledge standard, categorizing enterprise knowledge into five layers: raw documents, business topics, operation manuals, business rules, and Q&A analysis. This ensures that every response has a clear source, avoiding AI-generated content without basis, enhancing the reliability and compliance of enterprise knowledge management, while also supporting quick traceability and updates.

• End-to-End Trace and Feedback Loop: Record every sentence and each API call. Negative user feedback is automatically categorized and directed toward SOP modifications. Manual intervention is supported to refine experience back into the knowledge base, forming a continuous optimization cycle of "today's mistake, tomorrow's correction," preventing Agents from repeating errors and reducing long-term maintenance costs.

3. How to Use

1. Environment Requirements and Deployment: Access the GitHub repository (link to be updated after official release) with Ubuntu 20.04 or higher. A GPU with at least 8GB of VRAM (such as NVIDIA Tesla T4 or higher) is required to support model inference. Before deployment, install Docker and Docker Compose, and ensure that the network can access the model download source.

2. Creating a Digital Employee: Create a new employee profile in the management backend, setting the position, employee ID, capability boundaries, and department. This step is the foundation of employee-like governance, with each Agent corresponding to a formal employee ID, making it easy to integrate into the enterprise HR system for unified management. It is recommended to clearly define the position responsibilities during creation to facilitate the matching of corresponding SOPs and knowledge bases later.

3. Configuring Process Skills: Describe the business process in natural language, and the system will automatically distill it into a visual SOP state machine. After confirmation, publish it as a process-based skill, supporting multi-process chaining and interruption recovery. When configuring, pay attention to the boundary conditions of the process to avoid excessive state machine nodes that may complicate maintenance. It is recommended to start with simple processes for validation and gradually expand.

4. Building the Knowledge Base: Upload enterprise policy documents, and the system will automatically parse them in accordance with the OKF standard, establishing a structured, traceable knowledge asset. The knowledge base serves as the basis for digital employees to answer questions, ensuring accurate and reliable outputs. When uploading documents, it is recommended to use PDF or Markdown formats and categorize them by topic to improve parsing efficiency and accuracy.

5. Publishing and Operations: Publish the configured digital employee to the Employee Square for members within the organization to invoke based on their position needs. View the full-chain Trace through conversation logs, and continuously optimize SOPs and the knowledge base by combining automatic root cause analysis of user dissatisfaction with manual fallback mechanisms. During the operations phase, regularly review feedback data, adjust process nodes and knowledge entries, and establish a continuous iteration mechanism.

4. Pros and Cons Analysis

5. Comparative Analysis with Similar Tools

Selection Recommendations: For enterprises requiring strict employee governance, high process flexibility, and traceable knowledge, such as finance and government sectors with strong compliance requirements, StaffDeck stands out with its state machine SOP and OKF knowledge base, particularly suitable for organizations with well-established documentation systems and process standards. If an enterprise heavily relies on the Alibaba Cloud ecosystem and primarily needs remote operations and multimodal interaction, JVS Claw or Alibaba Cloud Digital Employee (Bailian) may be more advantageous in terms of integration convenience. For enterprises that have fully adopted Microsoft 365, Copilot Studio can seamlessly integrate into existing workflows, but it is relatively weaker in managing dedicated digital employee profiles and hierarchical knowledge traceability, making it more appropriate as an auxiliary tool rather than a core labor platform.

6. Editor's Summary

StaffDeck introduces an innovative concept in the field of enterprise-level AI Agent management known as "employee-like governance," elevating Agents from mere tools to organizational members—a perspective that remains relatively rare in the industry. Its state machine-driven workflow engine ensures the determinism of SOPs while incorporating flexible designs such as multi-process chaining and interruption recovery, making it more practically viable in engineering terms compared to pure large model planning or fixed workflow solutions. The hierarchical design of the OKF structured knowledge base fundamentally mitigates AI hallucination issues, enabling traceable and auditable responses, which is particularly crucial for highly regulated industries such as finance and healthcare.

In terms of practical value, StaffDeck lowers the threshold for enterprises to build digital employees: business personnel can generate SOPs using natural language without requiring programming skills. The automated root-cause analysis for negative feedback and the manual fallback mechanism create a sustainable optimization loop, reducing long-term maintenance burdens. The project utilizes the Apache 2.0 open-source license, with fully open code, allowing enterprises to perform secondary development on top of it and avoid vendor lock-in.

The target audience includes enterprise IT managers, process optimization teams, and organizations aiming to integrate AI Agents into their formal human resource systems. For technical teams, its microservices architecture and modular design facilitate integration and scalability. For non-technical users, the visual SOP configuration and knowledge base upload process are also relatively user-friendly.

In terms of future development potential, as the open-source community grows and more enterprise use cases accumulate, StaffDeck has the potential to become a benchmark project in the digital employee domain. The current version already covers core functionalities, and further improvements in documentation, reduction of deployment complexity, and expansion of pre-built workflow templates will make it more accessible for adoption in small and medium-sized enterprises. Additionally, deeper compatibility with mainstream large model platforms (such as OpenAI, Claude, etc.) would further expand its application boundaries.

7. Application Scenarios

• Expense Reimbursement: Digital employees collect invoices, verify standards, and submit for approval according to SOPs. If expenses exceed limits, the process is automatically escalated to human staff, and the entire process is traceable. Employees can submit reimbursement information through natural conversation, and the system automatically links to financial policies in the knowledge base, ensuring that every reimbursement complies with company regulations and reducing repetitive verification tasks for the finance department.

• IT Service Desk: Handles requests such as account permissions, device applications, and troubleshooting, assigning tickets according to SOPs. Complex issues are seamlessly escalated to human staff. Digital employees can quickly locate solutions for common problems using the IT manual in the knowledge base. For cases that cannot be handled automatically, the system generates a ticket and notifies the relevant engineer, improving the efficiency of IT service responses.

• HR Services: Answers frequent questions about leave policies, social security, and compensation benefits, providing clear and well-defined responses based on business rules. Employees can query policies using natural language, and the digital employee retrieves relevant clauses from the OKF knowledge base along with their sources, avoiding ambiguity in policy interpretation. It also supports transferring to HR staff for special cases requiring human intervention.

• Legal Compliance: Covers contract review and risk identification in clauses, providing review comments based on company compliance policies and historical cases. High-risk clauses are automatically escalated. Digital employees check contract elements item by item according to legal SOPs, combining compliance rules and case databases from the knowledge base to generate structured review reports, thereby enhancing the efficiency of the legal team.

• Administrative Affairs: Manages processes such as meeting room booking, office supply applications, and seal usage requests, standardizing routine tasks and freeing up the administrative team. Employees can complete applications through natural conversation, and the system automatically verifies resource availability and executes subsequent procedures. All operational records are traceable, facilitating resource allocation and auditing for the administrative team.

8. FAQ

Q: What hardware configuration is required for StaffDeck?
A: It is recommended to use a server equipped with an NVIDIA Tesla T4 or higher-performance GPU, with a minimum of 8GB of video memory. The CPU should support the AVX2 instruction set, and the memory should be at least 32GB. The recommended operating system is Ubuntu 20.04 or a later version, with Docker and Docker Compose installed. For lightweight testing only, StaffDeck can also run in a lower configuration environment, but the inference speed will be significantly reduced.

Q: Does StaffDeck support private deployment? How is data security ensured?
A: StaffDeck fully supports private deployment. All code and data are stored on the enterprise's local servers and do not rely on external cloud services. The platform provides full-chain log recording, allowing administrators to audit all conversations and API calls. The knowledge base and SOP data are stored in local databases, and enterprises can encrypt and back up data according to their own security policies.

Q: How can StaffDeck be integrated with existing enterprise systems (such as OA, ERP)?
A: StaffDeck provides standard RESTful API interfaces to enable data interaction with existing enterprise systems. In the SOP state machine, HTTP request nodes can be configured to achieve联动 (interoperability) with external systems. It also supports Webhook callbacks, making it easier to trigger subsequent business processes. For specific integration solutions, please refer to the API documentation in the official documentation.

Q: How is the knowledge base for digital employees updated? Does it support real-time updates?
A: The knowledge base is automatically parsed and built by uploading documents, and supports incremental updates. Administrators can upload new documents or modify existing knowledge entries at any time in the backend. After the update, digital employees will use the latest knowledge in subsequent conversations. For frequently changing business rules, it is recommended to regularly review and synchronize updates to maintain the accuracy of responses.

Q: What languages does StaffDeck support? Is it suitable for non-Chinese environments?
A: The current version is primarily optimized for Chinese scenarios, with the interface and documentation mainly in Chinese. However, the underlying model supports multilingual understanding, and theoretically, it can support English business processes by configuring an English knowledge base and SOP. For non-Chinese environments, users are advised to test and evaluate the effectiveness themselves. The community plans to add internationalization support in the future.

9. Project Links

• Project Website: https://staffdeck.openbmb.cn/
• GitHub Repository: https://github.com/OpenBMB/StaffDeck