如果说 2025 年是 Agent 元年，那么今年，无疑就是 Agent 的“爆发年”。
从写代码、做研究报告到处理日常业务，Agent 正在接手越来越多具体的工作。同时发生变化的，还有它与数据的关系。过去，数据库主要服务于人和业务系统，查询什么、如何调用，一般由开发者提前设计好。现在，Agent 也开始直接参与其中，成为数据使用者。但让 Agent 连上数据库，只是第一步。怎么让 Agent 快速找到需要的数据，并且知道这些数据该怎么用？背后还有不少问题需要解决。
这些变化，也让数据库厂商开始重新审视原有的数据架构。数据存储、检索、计算以及业务语义管理，都需要适应新的使用方式。
腾讯云今年 8 月推出的 AI 原生多模数据库 TDSQL Nexa Lakebase，就是在尝试将这些能力整合到一套统一的数据体系中。在 10 月 30 日举行的 2026 腾讯全球数字生态大会 AI 原生数据库专场上，腾讯云数据库进一步介绍了 TDSQL Nexa Lakebase 的技术架构与应用实践。
它试图回答的，是两个企业普遍关心的问题：企业数据攒了不少，却散在不同系统里，Agent 很难直接把它们组织起来用；就算找到了需要的数据，也未必读得懂它在业务上是什么意思。
Agent 能找到数据，就能真的用对吗？
第一个问题，是不少企业长期存在的老问题。只是到了 Agent 时代，随着 Agent 开始频繁调用不同来源的数据，原有的数据孤岛和系统协同问题变得更加突出。
在 Agent 出现以前，企业的业务数据就已经分散在不同系统中。比如，订单、交易记录等结构化数据通常存储在不同的关系型数据库中，文档、图片等非结构化数据可能保存在对象存储或文件系统里。每套系统都有自己的数据结构和处理方式，主要依靠接口和同步链路完成协作。
Agent 加入后，一个任务可能同时涉及结构化数据和非结构化数据，也可能需要在多个系统之间反复查询。如果继续沿用原来的架构，企业就需要处理更多数据同步、索引更新和结果整合工作。链路越多，暴露出来的问题也就越多。更麻烦的是，数据分散在不同系统中，Agent 首先得知道去哪里找。不同系统各有一套查询方式，检索结果还需要进一步关联和筛选。
即便找到了这些数据，Agent 也未必能真的用对。数据库中的表名、字段名和数据类型，只能描述数据的基本结构，很难完整表达企业内部的业务规则。比如，数据库中可能同时存在多个与“收入”相关的字段，但它们对应的业务口径可能不同。仅凭字段名称，Agent 很难判断应该使用哪一个。
这些隐藏在数据背后的字段含义、指标口径和业务规则，就是企业语义。Gartner 在今年发布的一项分析中指出，企业语义是影响 Agent 准确性的重要因素。Agent 需要理解企业数据之间的关系和规则，才能更准确地完成任务。
对企业来说，这两类问题往往交织在一起。Agent 要完成一次数据查询，既要找到正确的数据，也要弄清楚应该遵循什么业务口径。任务链条越长，跨系统调用越频繁，对底层数据组织和业务知识管理的要求也就越高。
TDSQL Nexa 如何重新组织多模数据与企业知识？
要让 Agent 准确高效地使用企业数据，数据库需要处理的不只是数据存储和查询，还包括多种数据形态的加工、检索，以及不同计算能力之间的协同。
腾讯云 TDSQL Nexa Lakebase 的思路，是把这些能力放进一套统一架构：融合宽表、搜索、分析和 AI 计算等引擎，减少数据在多个系统之间来回搬运；同时把表字段、业务术语、指标口径和专家经验整理成业务上下文，帮 Agent 理解数据背后的含义。
多引擎协同，减少数据在系统之间来回搬运
据腾讯云数据库副总经理、国产数据库中心负责人梁辰介绍，TDSQL Nexa 以对象存储 COS 作为数据底座，采用 Iceberg、Lance 等开放表格式承载海量原始数据和加工产物，并通过 GooseFS 提供分布式缓存加速，改善湖上数据的访问效率。
计算层采用可插拔的多引擎架构。宽表引擎负责海量特征、标签和业务数据的存储与查询；搜索引擎支持向量检索、全文检索和多条件过滤；分析引擎采用 MPP 分布式计算框架和向量化执行模式，处理大规模分析任务；AI 计算引擎则负责多模态数据加工，提供相应的 CPU、GPU 计算资源。
不同引擎负责各自擅长的任务，NTS 则负责连接这些引擎，让数据能够在不同处理环节之间流转。具体来说，NTS 既负责外部数据接入，将 MySQL、PostgreSQL、MongoDB 等数据源中的数据同步到 TDSQL Nexa。也负责内部引擎之间的数据同步，例如，宽表引擎中的数据可以通过 NTS 同步到搜索和分析引擎，供后续的向量检索、全文检索和数据分析使用。这样，同一份业务数据就能支持不同类型的查询，不必由应用层反复拼接结果。
如果企业暂时不希望迁移原有数据，也可以通过 ShortCut 联邦访问外部数据源，直接查询原系统中的数据。企业可以根据数据实时性、性能和管理需求，选择原地访问或同步纳管。目前，TDSQL Nexa 支持接入 MySQL、PostgreSQL、MongoDB、COS/S3 等多类数据源。
在不同引擎已经接通的基础上，TDSQL Nexa 还需要解决一个问题：开发者和 Agent 如何使用这些能力？
TDSQL Nexa 在上层设计了统一的智能编排路由引擎，通过声明式 DDL 定义逻辑表、字段类型和索引类型。查询发生时，系统会根据查询需求和代价模型，判断调用哪个引擎、使用哪种索引完成查询。用户不必分别了解底层各个引擎的数据组织方式。对外，TDSQL Nexa 提供 SQL、REST、MCP 等访问方式，计算侧兼容 MySQL、PostgreSQL、OpenSearch 等协议生态，降低现有系统和开发工具的接入改造成本。
为 TDSQL Nexa 提供多模数据底座的，是腾讯云数据库 TDSQL Boundless。它在原有分布式事务与可扩展存储的能力上，扩展了自动冷热分离的功能来降低系统使用成本，同时构建了宽表、全文、向量和列存能力。
对于 AI 应用产生的大量数据，Boundless 可以结合数据访问情况进行冷热分层：热数据保留在本地 CBS，冷数据自动下沉到 COS。针对占用空间较大的字段，例如 Agent 对话内容和图片等，也可以将其放到对象存储中，其他元数据继续保留在本地，以兼顾访问性能和存储成本。
AI 计算引擎：让多模态数据加工与后续使用衔接起来
多模态数据进入系统后，还需要经过解析、抽帧、打标和向量化等处理，才能用于后续检索和训练。在传统架构中，这些工作往往由不同的计算任务和系统分别承担，加工完成后还需要将结果同步到数据库或搜索引擎。
TDSQL Nexa 将 AI 计算引擎 纳入整体架构，统一纳管 CPU、GPU 异构算力，并提供数据处理、分布式计算和作业编排能力。图片、视频等原始数据经过解析、质量过滤、推理打标和 Embedding 生成后，产生的向量、标签和元数据可以进一步与宽表、搜索等引擎衔接，供后续检索和分析使用。
以训练数据准备为例，算法团队需要从大量原始数据中筛选有价值的样本，再进行特征提取和数据加工。AI 计算引擎可以承接其中的处理任务，并将加工结果接入后续的数据存储与检索链路，减少不同系统之间的数据搬运和重复加工。
Nexa Knowledge：让 Agent 理解企业自己的业务语言
把不同数据放进一套架构，让 Agent 能够统一查询，只解决了问题的一部分。
正如前文所说，Agent 即使找到了正确的数据，也可能因为不了解企业内部的业务规则而得出错误结果。数据库的表结构和字段信息，并不能完整表达这些知识。为此，TDSQL Nexa 还在数据引擎之上增加了 Knowledge 能力，用来构建和维护企业语义。
Nexa Knowledge 先从已有数据库中提取表、字段和关联关系，建立基础的数据语义；企业还可以补充指标定义、业务术语、业务规则、文档和专家经验。 系统对这些材料进行解析和整理，将有价值的信息沉淀为 Skill、术语、澄清规则、已验证问答和世界事实五类知识资产。
Agent 接到查询任务时，可以检索相关知识，了解应该使用哪些数据、遵循什么指标口径，以及采用怎样的查询方法。这些信息可以作为 SQL 生成和数据分析的依据，减少模型仅凭字段名称推测查询逻辑的情况。
这套语义体系还会随着业务使用持续更新。新的文档、字段变化，以及查询过程中发现并确认的语义偏差，都可以重新沉淀到对应的知识资产中，供后续任务复用。对企业来说，原本散落在数据结构、业务文档和专家经验中的知识，终于可以逐步形成一套能够被 Agent 检索和调用的语义体系。
从多引擎协同到 AI 计算引擎，再到 Nexa Knowledge，TDSQL Nexa 试图把数据存储、处理和业务语义管理连接起来。数据可以在不同引擎之间流转，企业积累的业务知识也有了统一沉淀和持续更新的地方。最终，让 Agent 更准确地找到、理解和使用企业数据。
从智能驾驶到企业 Agent，新的数据架构如何落地？
一套新的数据架构有没有价值，最终还是要看它能否解决实际业务问题。腾讯云数据库在大会上分享了两个基于 TDSQL Nexa 的客户案例，一个来自智能驾驶，另一个来自教育行业。
先看智能驾驶，这是多模态数据处理需求比较典型的场景。
一辆车在行驶过程中会产生大量图像、视频、传感器信号和事件日志。算法团队需要从这些数据中找到有价值的训练样本，尤其是一些出现概率较低、但对模型能力提升非常重要的长尾场景。
以车辆紧急制动为例，车端会将事件前后一段时间的数据打包成 Lance 文件，上传到对象存储；事件和文件的元数据存放在 MongoDB，合规和地图数据则保存在 PostgreSQL。随后，基于 Spark 或 Ray 的计算管线会提取相关特征，包括天气、路况、行人和车辆的位置、速度等信息。一次事件可能产生数百列特征描述，累计特征维度达到上万。算法团队需要从历史数据中寻找相似场景，筛选样本用于模型增量训练。
原有架构中，Doris 负责存储标签数据，ES 负责存储语义向量。不同系统各自管理数据和索引，当算法团队需要同时按照语义相似度、天气、道路和车辆状态等条件筛选样本时，还需要在应用层融合不同系统的查询结果。
引入 TDSQL Nexa 后，数据处理链路得到简化：原始数据上传至对象存储，AI 计算引擎负责特征提取和加工，处理后的特征数据写入宽表引擎，再通过 NTS 同步到搜索和分析引擎，更新相应索引。算法团队筛选样本时，可以通过一条 SQL 完成全文、向量和标量条件的混合检索，不必再分别访问多个系统并在应用层拼接结果。据介绍，该车企改造后的数据闭环效率提升了约 10 倍。
第二个案例来自教育行业。这家企业希望通过 Agent 实现对话记忆、课程资料检索，以及课程和题目相关的问题解答。实际建设中，需要管理的数据包括 Agent 会话、个人信息记录、课件和课程视频等，还涉及相应的数据加工、搜索和分析系统。原有架构中，这些能力分散在不同系统，维护成本较高，Agent 回答准确率也受到影响。
基于 TDSQL Nexa，企业将数据存储、加工、检索和分析整合到统一的数据体系中。与此同时，Nexa Knowledge 对数据库中的表、字段及其关系进行梳理，并将课程文档、视频内容和业务规则纳入企业知识体系，让 Agent 能够结合数据和业务上下文回答问题。
随着新文档持续写入、数据库字段发生变化，相关语义也可以随之调整；使用过程中发现的语义偏差，则可以经过校验后重新沉淀到知识体系中。据介绍，在综合评测中，这套方案的问题回答准确性提升了 60% 以上。
数据库的边界正在延伸
Agent 对数据库的影响，已经不只是查询方式发生变化。据腾讯云披露，过去一年，Agent 创建的腾讯云数据库（TDSQL-C）实例日均超过 2 万个，增速超过 80 倍，活跃实例突破 100 万个。Agent 已经开始大量参与数据库的创建和使用，数据库需要适应的工作方式，也需要发生变化。
腾讯云副总裁王义成在此前接受采访时，曾将这种变化概括为 Agent 带来的“未知”。Agent 由模型生成，什么时候发起请求、会产生什么样的数据、给数据库带来多大的压力，都很难按照传统业务的方式提前预估。数据库需要面对的，除了新的数据形态，还有更加难以预测的访问负载和运行需求。
从腾讯云面向 Agent 的数据库产品布局来看，相关能力已经延伸到多个方向：TDSQL-C 面向 Agent 应用后端开发，具备 Branch 秒级分支、Serverless 弹性、多 Agent 运行沙箱三大特性；TencentDB Agent Memory 负责记忆与团队知识管理，自 5 月开源以后多次登上 GitHub Trending 全球第一，目前收获了超 3w Stars；TencentDB Agent 则是全新的数据库智能体，具备感知、分析、决策、执行的完整自治能力。
当 Agent 开始承担越来越多具体工作，数据库的边界将不断延伸，既要保证数据安全、稳定和可用，也要让 Agent 能够找到数据、理解数据，并在业务过程中正确使用数据。这些能力能否真正经受住生产环境的考验，才是 AI 原生数据库接下来需要回答的问题。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
If 2025 is Agent's year, then this is undoubtedly Agent's “year of outbreak”.
Agent is taking over a growing number of specific tasks, from writing codes, doing studies and processing routines. At the same time, there are changes in its relation to data. In the past, databases used to serve primarily human and operational systems, search for what and how to call, and are generally designed in advance by developers. Now, Agent is also directly involved, becoming a data user. But getting Agent into the database is just the first step. How can Agent quickly find the required data and know how they work? There are many issues that need to be addressed.
These changes have also allowed database manufacturers to begin revisiting the original data architecture. Data storage, retrieval, computing and business semantic management need to be adapted to new uses.
The Tencent cloud launched in August of this year's AI multi-modular database TDSQL Nexa Lakebase is an attempt to integrate these capabilities into a unified data system. The technical architecture and application of TDSQL Nexa Lakebase was further described in the Tencent Global Digital Ecology Conference, held on 30 October, dedicated to the AI Native Database.
It seeks to answer the general concern of two enterprises: the fact that business data are abundant but scattered across different systems, and Agent has difficulty organizing them directly; even if the required data are found, it does not necessarily understand what it means to operate.
Agent can find the data and actually use it?
The first is the long-standing problem of many enterprises. It's just that, in the Age of Age, the issue of old data island and system synergy became more prominent as Agent began to make frequent calls from different sources.
Before Agent appeared, business data for enterprises were scattered across different systems. For example, structured data, such as orders, transaction records, are usually stored in different relationship databases, and unstructured data, such as documents, pictures, may be stored in object storage or file systems. Each system has its own data structure and processing, relying primarily on interfaces and synchronized links for collaboration.
With the addition of Agent, a task may involve both structured and non-structured data and may require repeated searching between multiple systems. If the original structure is to continue, enterprises will need to process more data synchronization, index updating and integration of results. The more links, the more problems are exposed. Even more troublesome is that data is scattered across different systems, and Agent has to know where to look first. There is a set of searching methods in different systems and the results need to be further linked and screened.
Even if these data are found, Agent may not really be able to use them. Table names, field names and data types in the database can only describe the basic structure of the data and it is difficult to express the business rules within the enterprise. For example, multiple fields associated with “income” may exist simultaneously in the database, but they may have different business calibres. Only field names make it difficult for Agent to judge which one should be used.
The meaning of field, the calibration of indicators and the rules of operation, which are hidden behind the data, are corporate syntax. Gartner, in an analysis this year of publish, pointed out that business syntax is an important factor affecting the accuracy of Agent. Agent needs an understanding of the relationship and rules of business data in order to be able to fulfil its mandate more accurately.
For enterprises, these two types of issues are often intertwined. Agent had to complete a data search to find the right data and to see what business calibre should be followed. The longer the task chain, the more frequent cross-system calls, the higher the demand for bottom data organization and business knowledge management.
How can TDSQL Nexa reorganize multi-mode data and business knowledge?
To enable Agent to use business data accurately and efficiently, the database needs to deal not only with data storage and searching, but also with processing, retrieving and synergizing different data forms.
The idea behind Tencent cloud TDSQL Nexa Lakebase is to place these capabilities in a unified set of structures: integration of broad meters, search, analysis and AI computing engines to reduce data moving back and forth between multiple systems; and to organize table field, business terminology, indicator calibres and expert experience into business context to help Agent understand the meaning behind the data.
Multi-engine synergies. Reduce data moving back and forth between systems. Transport
According to Liang Jie, Deputy General Manager of the Tencent Cloud Database and Head of the National Production Database Centre, TDSQL Nexa uses object storage COS as the data base, loads bulk raw data and process products in open-table formats such as Iceberg and Lance, and provides distributed cache access efficiency through GooseFS.
The computational layer uses a multi-engine structure that can be inserted. The broad-table engine is responsible for the storage and searching of mass features, labels and operational data; the search engine supports vector retrieval, full-text retrieval and multi-condition filtering; the analytical engine handles large-scale analytical tasks using the MPP distributed computing framework and the quantization model; and the AI computing engine is responsible for the processing of multi-modular data, providing corresponding CPU, GPU computing resources.
Different engines are responsible for their own tasks, while NTS is responsible for connecting these engines so that data can flow between the different processes. Specifically, NTS is responsible for external data access by synchronizing data from data sources MySQL, PostgreSQL, MongoDB to TDSQL Nexa. It is also responsible for data synchronization between internal engines, e.g., data in broad-scale engines can be synchronized through NTS to search and analysis engines for subsequent vector retrieval, full-text retrieval and data analysis. In this way, the same business data can support different types of queries without having to be collusived at the application level.
If an enterprise does not wish to move the original data for the time being, it can also access external data sources through the ShortCut Federation and directly search the data in the original system. Depending on the real-time nature of the data, performance and management needs, an enterprise may choose to access or synchronize the tube in situ. Currently, TDSQL Nexa supports access to multiple types of data sources such as MySQL, PostgreSQL, MongoDB, COS/S3.
On the basis that different engines have been connected, TDSQL Nexa also needs to address the question: how do developers and Agent use these capabilities?
TDSQL Nexa has designed a unified intelligent router engine at the top level to define the logical table, field type and index type through a declarative DDL. When a query occurs, the system will determine which engine to call and which index to use to complete the query, based on the search demand and cost model. Users do not need to know how the data are organized separately for the bottom engines. Externally, TDSQL Nexa provides access to SQL, REST, MCP etc. to calculate the ecological aspects of protocols such as MySQL, PostgreSQL, OpenSearch, etc., to reduce the cost of accessing and adapting existing systems and development tools.
The multi-molecular data base for TDSQL Nexa is the Tencent cloud database TDSQL Bonnles. It expands the automated cold-heat separation function to reduce system usage costs in terms of existing distributed services and scalable storage capabilities, while building broad-scale, full-text, vector and memory capabilities.
For the large amount of data generated by AI applications, Boundless can make cold thermal layers in conjunction with data access: thermal data are kept in local CBS and cold data automatically sink to COS. For the larger space occupied by field, e.g. Agent dialogue content and pictures, they can also be stored in object storage, while other metadata remain locally to balance access performance and storage costs.
AI Calculator Engine: Linking multimodular data processing to subsequent use
After entering the system, multi-modular data will also need to be processed through decomposition, frames, marking and to quantization for subsequent retrieval and training. In traditional structures, these tasks are often carried out separately by different computing tasks and systems, and after processing they also need to synchronize the results to databases or search engines.
TDSQL Nexa integrates the AI computing engine into the overall architecture, harmonizes the CPU, GPU isomer algorithms and provides data processing, distributed computing and operation organization capabilities. Raw data such as pictures, videos, etc., are analyzed, quality filtered, reasoned and Embedding generated, resulting vectors, labels and metadata can be further linked to broad-scale, search engines for subsequent search and analysis.
In the case of training data preparation, the algorithm team would need to screen valuable sample from a large volume of raw data for characterization and data processing. The AI computing engine can take over the processing tasks and access the results to subsequent data storage and retrieval links, reducing data handling and duplicate processing between different systems.
Nexa Knowledge: Let Agent understand the business's own business language
Puts different data in a set of structures that allows Agent to search in a uniform way, only partially solves the problem.
As stated earlier, even if the correct data were found, Agent could have made the wrong result by not knowing the business rules within the enterprise. The table structure of the database and the field information do not fully convey this knowledge. To this end, TDSQL Nexa has added Knowledge capabilities to the data engine to build and maintain business semantics.
Nexa Knowledge first extracts tables, field and links from existing databases and establishes basic data syntax; enterprises can also add definitions of indicators, business terminology, business rules, documentation and expert experience. The system interprets and collates these materials, depositing valuable information into Skill, terminology, clarification of rules, validated questions and answers, and five types of knowledge assets of world fact.
Upon receiving a search mission, Agent is able to retrieve knowledge about what data should be used, what calibres of indicators should be followed and what type of search methods should be used. This information can be used as a basis for SQL generation and data analysis to reduce the number of cases where models use field only to speculate on search logic.
The semantic system will be updated as operations are used. New documents, field changes, and semantic deviations identified and confirmed during queries can be repositioned to corresponding knowledge assets for reuse in subsequent assignments. For enterprises, knowledge that was scattered over data structures, business files and expert experience can eventually evolve into a semantic system that can be retrieved and adapted by Agent.
From multiple engines to the AI computing engine, to Nexa Knowledge, TDSQL Nexa attempts to connect data storage, processing and business semantic management. Data can flow between different engines, and business knowledge accumulated by enterprises can be consolidated and continuously updated. Eventually, Agent will find, understand and use business data more accurately.
How does the new data structure land from smart driving to business, Agent?
The value of a new data architecture will ultimately depend on its ability to address operational issues. The Tencent Cloud Database shared at the Conference two client cases based on TDSQL Nexa, one from smart driving and the other from the education industry.
It's a typical scenario for multi-modular data processing needs.
A vehicle produces a large number of images, videos, sensor signals and incident logs during its journey. The algorithm team needs to find valuable training sample from these data, in particular long tailings that are less probable but are important for model capacity enhancement.
In the case of emergency vehicle brakes, the end of the vehicle will package data for a period of time before and after the event into a Lance file and upload it to the object for storage; metadata on events and documents will be stored in MongoDB and compliance and map data will be stored in PostgreSQL. Subsequently, computational tubes based on Spark or Ray extract relevant features, including weather, road conditions, location of pedestrians and vehicles, speed, etc. An event may produce hundreds of characterizations, with cumulative characterization dimensions of tens of thousands. Algorithm teams need to look for similar scenarios from historical data and screen sample for model incremental training.
In the old architecture, Doris stored tag data, and ES stored semantic vectors. Different systems manage data and indexes separately, and when an algorithm team is required to screen both semantic similarity, weather, road and vehicle status When selecting sample, the results of the query need to be integrated into different systems at the application level.
The introduction of TDSQL Nexa has simplified the data-processing links: raw data are uploaded to object storage, the AI computing engine is responsible for characterization and processing, processed feature data are written into the broad-table engine, and the corresponding index is updated via NTS sync to the search and analysis engine. When the algorithm team screens sample, a combination of full text, vector and cursor conditions can be completed through a SQL without separate access to multiple systems and collating results at the application level. It was described that the data closure efficiency of the vehicle had increased about 10 times.
The second case comes from the education sector. This enterprise hopes to achieve dialogue memory, course information retrieval, and answers to questions about courses and topics through Agent. Among the data to be managed in practice are Agent sessions, personal information records, course packages and course videos, and the corresponding data processing, search and analysis systems. In the original structure, these capabilities were dispersed across different systems, with high maintenance costs and Agent's response accuracy being affected.
Based on TDSQL Nexa, enterprises integrate data storage, processing, retrieval and analysis into a unified data system. At the same time, Nexa Knowledge combed tables, field and its relationships in the database and incorporated course files, video content and business rules into the enterprise knowledge system, allowing Agent to answer questions in combination with data and business context.
As the new document continues to be written and the database field changes, the semantics can be adjusted accordingly; the semantic deviations observed during use can be re-calibrated into the knowledge system after validation. It was explained that the accuracy of the responses to the questions of the package had increased by more than 60 per cent in the comprehensive assessment.
Database boundaries are being extended
Agent's influence on the database has gone beyond simply changing the search mode. According to the Tencent cloud, the Tencent cloud database (TDSQL-C) created by Agent has exceeded 20,000 examples a day over the past year, with an increase of more than 80 times, and active cases have exceeded 1 million. Agent is already heavily involved in the creation and use of the database, which needs to be adapted to the working methods and changes.
In an interview with Vice President Tencent.K. Wang Yi, the change was summarized as “unknown” by Agent. It is difficult to predict ahead of schedule in the traditional way of doing business when it is generated by models, when requests are launched, what data will be generated and how much pressure will be placed on databases. The database needs to be faced with more unpredictable access loads and operational requirements in addition to new data patterns.
From the layout of the database products of the Tencent cloud towards Agent, the capacity has been extended to multiple directions: TDSQL-C, which is developed at the back end of the Agent application, has three main characteristics of the Branch Second-Spectrometer branch, Serverless, MultiAgent Run Sandbox; TencentDB Agent Memoory, which is responsible for memory and team knowledge management, has been on the GitHub Trending Global number one many times since the start of May and is now reaping over 3w Stars; TencentDB Agent is an entirely new database AI agent with full self-government in perception, analysis, decision-making and implementation.
As Agent begins to take on more and more specific tasks, the boundaries of the database will continue to extend, both to ensure data security, stability and availability, and to enable Agent to find, understand and use data correctly in its operations. Whether or not these capabilities really stand the test of the production environment is the question that the AI primary database needs to answer.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.