zj-unicom-ai/UniEmployee

What it solves

UniEmployee is an enterprise-grade platform designed to transform professional business workflows into autonomous digital employees. Instead of being a generic chatbot, it allows organizations to codify specific business processes, judgment criteria, and operational skills into configurable AI agents capable of handling roles like customer service, data analysis, HR, and network operations.

How it works

The platform uses a five-layer capability model (Employee → Workflow/SOP → Skill → Connector → Tool) to manage AI agents. Administrators configure agents via a web interface, defining their persona, knowledge base, and specific operational procedures (SOPs). The system supports long-term memory, state-machine workflows for complex tasks, and "Human-in-the-loop" (HITL) approvals for high-risk actions. It integrates with external systems via MCP (Model Context Protocol) and provides full observability, allowing users to trace every step of an agent's reasoning, tool usage, and data retrieval.

Who it’s for

This project is for enterprises and organizations looking to build and deploy specialized AI agents that follow strict business logic rather than relying on general-purpose LLM responses. It is suited for technical teams or business analysts who want to integrate AI into existing data pipelines, CRM systems, and internal knowledge bases while maintaining security and auditability.

Highlights

• Structured Knowledge Ontology: Moves beyond simple RAG by organizing business knowledge into entities, relationships, and rules, allowing agents to perform multi-hop reasoning.
• Process-Oriented Skills: Uses SKILL.md files and state machines to ensure agents follow defined Standard Operating Procedures (SOPs) rather than improvising.
• Full Observability: Every interaction is recorded as a 'run,' providing a detailed trace of LLM inputs, tool calls, and execution time for debugging and auditing.
• Security & Guardrails: Includes SQL read-only validation, tool whitelisting, sensitive word filtering, and human-in-the-loop approval nodes to prevent unauthorized actions.
• Data Analysis Workbench: Features a dedicated analyst workspace with SQL safety checks and report generation capabilities.