对以“研究”为主业的智库机构而言，最宝贵的是什么？
答案是研究员的精力。
然而，他们的精力往往消耗在翻查政策库、核对口径、调整格式、整理材料这些重复性琐事上。
近期，国家数据局直属的国家数据发展研究院正式启用WorkBuddy专享版，并与腾讯乐享知识库配合使用，服务院务办公与技术研究两大板块。
作为全国首个落地国家级数据智库的AI办公智能体，WorkBuddy解决“会干活”的问题，承担科研任务的理解、拆解与执行；腾讯乐享知识库则充当知识底座，保证智能体干活“有依据”。
把科研院务中的重复环节交给AI
作为典型的知识密集型组织，国家数研院等科研机构的核心竞争力在于研究能力，而非资料搬运能力。
WorkBuddy“入职”的初衷很朴素：把研究员从检索、摘录、比对、格式整理这些必要但低价值的环节中解放出来，让时间和精力真正投向判断与洞见。
首批应用聚焦院务办公与技术研究两个方向，覆盖四大场景：
院务办公方面：
- 公文大模型：主责“写”与“审”，承担政策及院务公文的智能起草、合规校对与自动归档，AI负责查找、整理和生成，人负责审核与判断；
- 科研院务超级助理：主打办事，把会议、督办、报销、资产等事务收进一个对话入口，实现事项对话式办理，一句话即可完成报销单据识别与预填等工作；
- 智能知识中枢：智能化存储和调用数研院内部知识，依托腾讯乐享知识库，把院内课题、标准、报告等知识资产统一纳管，并根据使用反馈持续更新，用大白话提问，即可获得逐一标明出处、可回溯原文的回答。
技术研究方面：
- AI研究员助手：贯穿文献研读、仿真计算、实验设计全流程，既能批量解析中外文献、提炼观点方法与结论并形成对比综述，也能辅助搭建模型、运行参数、生成图表。
除此之外，各业务部门还能按自身需求定制专属技能，把常用流程与研究经验固化下来，沉淀为可复用的能力。
专属版本，研究数据全程可管可控
研究国家数据战略的智库机构使用AI，数据安全与合规没有商量余地。
此次落地国家数研院，WorkBuddy专享版以三重保障确保数据安全合规：
- 专属空间、数据隔离：运行于腾讯云为数研院独立开通的专属环境，资源独占、租户隔离、数据独立存储，院内数据与外部环境物理隔离，未公开的研究底稿、论证材料与内部数据全程可管可控；
- 统一管理、权限到人：通过企业管理后台统一管理成员账号、组织架构与可用范围；依托腾讯乐享知识库的四级权限体系，AI无论执行任务还是调用知识，都严格限定在使用者原有岗位权限之内，既“用得上”又“管得住”；
- 沙箱执行、操作留痕：每项任务默认在安全沙箱中运行，操作全程留痕、风险行为实时拦截，为追溯与审计提供依据。
腾讯云副总裁、政务政法行业负责人胡鑫表示，研究机构、智库、高校乃至各级政府的政策研究部门，都面临知识资产沉睡、专家经验难沉淀、研究流程难标准化的共同困境。
未来，我们将与国家数研院携手，进一步深化智能体在科研场景中的应用，让国家级智库的先行探索，沉淀为更多机构可复用的路径。
本文来源：https://www.163.com/dy/article/L6D3PG9P0556H003.html
若有版权问题，请来信告知：13951431913#139.com 或在右方提交评论。
For think tanks whose primary occupation is "research," what is the most precious resource?
The answer is the energy of the researchers.
However, their energy is often consumed by repetitive tasks such as searching policy databases, checking data consistency, adjusting formats, and organizing materials.
Recently, the National Institute of Data Development, directly affiliated with the National Data Administration, officially adopted the WorkBuddy exclusive version and combined it with the Tencent Leos Knowledge Base to serve the two sectors of institute administration and technical research.
As the first AI office agent to be implemented in a national-level data think tank, WorkBuddy solves the problem of "doing the work" by undertaking the understanding, breakdown, and execution of research tasks; while the Tencent Leos Knowledge Base acts as the knowledge foundation, ensuring that the agent performs its duties with "evidence."
Handing over the repetitive steps of scientific research and administration to AI
As a typical knowledge-intensive organization, the core competitiveness of research institutions like the National Institute of Data Development lies in their research capabilities, not in data retrieval and filing.
The initial purpose of introducing WorkBuddy was simple: to free researchers from low-value, necessary tasks such as searching, extracting, comparing, and formatting, so that time and energy could be directed toward judgment and insight.
The first batch of applications focused on two directions — institute administration and technical research — covering four scenarios:
In terms of institute administration:
Document Large Model: Responsible for "writing" and "reviewing," it undertakes the intelligent drafting, compliance checking, and automatic archiving of policy and administrative documents. AI handles the finding, organizing, and generating, while humans perform the reviewing and judging.
Scientific Research Administrative Super Assistant: Specializing in executing tasks, it consolidates meetings, follow-ups, reimbursements, and asset management into a single conversational interface, enabling dialogue-based processing of matters. For example, one can complete the recognition and pre-filling of reimbursement slips with a single sentence.
Intelligent Knowledge Hub: Provides intelligent storage and retrieval of internal knowledge. Relying on the Tencent Leos Knowledge Base, it unifies the management of internal knowledge assets such as projects, standards, and reports, and continuously updates based on user feedback. This allows for asking questions in plain language and receiving responses that are clearly cited and traceable to the original source.
In terms of technical research:
AI Researcher Assistant: Spans the entire process of literature review, simulation calculation, and experimental design. It can not only batch-parse Chinese and foreign literature to extract viewpoints, methods, and conclusions into comparative summaries, but also assist in building models, running parameters, and generating charts.
Additionally, various business departments can customize exclusive skills according to their own needs, solidifying common processes and research experiences into reusable capabilities.
An exclusive version with fully manageable and controllable research data
For a think tank researching national data strategies, there is no room for negotiation regarding data security and compliance.
The implementation of the WorkBuddy exclusive version at the National Institute of Data Development ensures data security and compliance through three layers of protection:
Exclusive space and data isolation: It operates within an independent cloud environment exclusively opened for the institute by Tencent Cloud. The resources are exclusive to the tenant, data is stored independently, and the internal data of the institute is physically isolated from the external environment — meaning unpublished research drafts, substantiation materials, and internal data are fully manageable and controllable.
Uniform management with person-specific permissions: Member accounts, organizational structures, and available ranges are managed uniformly via the enterprise management backend. Based on a four-level permission system, AI is strictly limited to the user's existing job permissions, regardless of whether it is executing a task or retrieving knowledge.
Sandbox execution and traceable operations: Every task is run in a safe sandbox by default, with all operations logged and risk behaviors intercepted in real-time, providing a basis for tracing and auditing.
Hu Xin, Vice President of Tencent Cloud and Director of the Government and Public Security Industry, stated that research institutions, think tanks, universities, and policy research departments at various levels of government all face the common problem of sleeping knowledge assets, difficult-to-settle expert experiences, and non-standardizable research processes.
In the future, we will partner with the National Institute of Data Development to further deepen the application of intelligent agents in research scenarios, using the pioneering exploration of national-level think tanks as a reusable path for other institutions.