WPS 365 Launches Official AI Little Lobster

🔑 Enhanced Key Takeaways

• •Xiao K (Little Lobster) serves as the unified interface for Kingsoft's 'Digital Employee 2.0' framework, transitioning from a reactive chatbot to an autonomous agent capable of cross-application task orchestration between IM, Docs, and Meetings.
• •The 'result accountability' feature is anchored by the 'Qingzhou' engine, a native Xinchuang-compliant infrastructure that provides deterministic execution layers and verification audit trails to mitigate AI hallucinations in professional documents.
• •The platform utilizes a 'Model-as-a-Service' (MaaS) architecture, allowing enterprises to toggle between multiple domestic LLMs (such as MiniMax, Zhipu AI, and Baichuan) based on specific task requirements or cost constraints.
• •WPS 365 has introduced a 'Private-Public Hybrid' deployment model specifically for Xiao K, ensuring that sensitive enterprise knowledge remains on-premises while leveraging public cloud LLMs for high-level reasoning tasks.

🛠️ Technical Deep Dive

• •MaaS (Model-as-a-Service) Architecture: Decouples the application layer from the underlying LLM, enabling seamless switching between different models for text, code, or image generation.
• •Knowledge-Enhanced Generation (RAG): Implements a proprietary retrieval system that indexes unstructured enterprise data (PDFs, chats, spreadsheets) to provide context-aware responses with source citations.
• •Qingzhou Engine: A self-developed, native engine designed for compatibility with domestic Chinese operating systems (HarmonyOS, Kylin) and hardware (Kunpeng, Loongson).
• •Multi-Agent Orchestration: Uses a planning layer to decompose complex user prompts (e.g., 'Analyze this meeting and draft a contract') into sub-tasks executed across different WPS components.
• •Traceability Engine: A security layer that logs AI decision-making paths, allowing human supervisors to audit how the AI arrived at a specific document draft or data analysis.

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline