Kingsoft Office Launches Standalone AI-Native Office Agent 'Lingxi Pro'

Kingsoft Office released Lingxi Pro at the 2026 AI Productivity Conference. It is an independent AI-native office agent, not an embedded feature of WPS. Its core logic is 'not adding AI to Office, but making AI grow Office.' This means Lingxi Pro and WPS work in division: WPS handles deep editing and formal circulation, while Lingxi Pro organizes projects, mobilizes materials, executes tasks, and delivers results.

Lingxi Pro connects to multiple mainstream domestic large models. Kingsoft Office conducts scenario-specific adaptation tests on different models, automatically matching the optimal model for different tasks, with seamless user switching. This engineering adaptation enables the agent to have a longer delivery chain in task understanding, file generation, rendering verification, etc.

The key difference of Lingxi Pro lies in its file generation capability. It calls WPS CLI—a document operation kernel open to agents—that can truly create, modify, and generate standard Office native objects. In generated PPTs, charts are independent image elements, tables are editable table objects, charts have data sources, and text is modifiable text placeholders. All elements comply with Office standards and can be directly edited.

In spreadsheet scenarios, tables generated by Lingxi Pro fully retain formulas, conditional formatting, and data relationships. When users modify original data, calculation results and associated charts update accordingly, indistinguishable from manually built tables. Before delivery, AI automatically performs rendering verification, identifying and fixing issues like misalignment, occlusion, and layout errors.

In contrast, most AI office products on the market generate documents that are essentially 'pseudo-files': tables are static numbers, charts are images, layouts are stacked text boxes, requiring extensive rework. Lingxi Pro overcomes this industry challenge through engineering capability.

Lingxi Pro organizes work context by 'project.' Conversations, documents, materials, and progress within the same project are all deposited in a unified space. When users assign tasks, the agent automatically matches existing materials and supplements with internet searches and third-party data sources as needed, understanding the background throughout. The official statement: 'Your AI doesn't need to be re-onboarded every day.'

The 'Auto Dream' function is a background cognitive and memory management mechanism that automatically summarizes the day's conversations and files during idle compute time, performing information summarization, compression, and structured organization. The memory system has three layers: core memory (personal information, writing style, collaboration rules), knowledge base (project materials, industry documents), and auto dream (cross-session summarization). All memory is based on user authorization, fully visible, modifiable, and can be turned off.

In 2026, AI office enters the 3.0 agent era: users set goals and constraints, AI autonomously breaks down tasks, calls tools, advances processes, and delivers usable results. Overseas and domestic leading vendors have released products intensively, such as Anthropic's Claude Cowork, Tencent's Workbuddy, and OpenAI's ChatGPT Work. Within a week of ChatGPT Work's launch, combined active users of Codex and ChatGPT Work exceeded 8 million, and platform token usage surged 2.5 times.

However, all players face a common challenge: it's easy to run an agent, but hard to deliver usable results. The vast majority of AI office products generate documents that are essentially 'pseudo-files' that cannot be directly edited. This hurdle cannot be overcome by stacking model parameters; it requires engineering capability. Kingsoft Office's engineering barriers—document kernel, layout engine, rendering verification—are the result of decades of accumulation, difficult for general large model vendors to replicate in the short term.

Kingsoft Office CEO Zhang Qingyuan proposes that office software is moving from standalone office and collaborative office to AI office. 'AI will reshape the paradigm of work completion: in the past, humans operated software; today, humans set goals and AI participates in completion.' Kingsoft Office believes that large model technology will eventually reach parity, and what truly determines product competitiveness is the depth of understanding of user context and the office data assets accumulated in real scenarios.

Kingsoft Office Assistant President Tian Ran proposed three hard standards for measuring AI office assistants: understanding user context, completing tasks with high quality, and possessing professional Office operation capabilities. The first two rely on basic large model capability iteration and engineering orchestration, while the latter demonstrates the gap from long-term business accumulation. Kingsoft Office simultaneously launched the C-end Lingxi Pro and the B-end WPS Comate, targeting personal scenarios and enterprise markets respectively.