CodeBuddy IDE

• 1. overview (989 chars)

  Tencent CodeBuddy is an AI programming assistant officially launched by Tencent Cloud in May 2024. Based on Tencent's Hunyuan code large model and DeepSeek dual engine driver, it provides full-process AI programming assistance capabilities for enterprise teams and individual developers. This product is one of the few AI programming tools in China and even in the world that supports three product forms: plug-in, independent IDE and CLI. It focuses on the concept of "conversation is programming" and the integrated development experience of industry, equipment and research. As of 2025, more than 90% of Tencent's engineers are using CodeBuddy, AI-generated code accounts for more than 50%, and the overall coding time has been shortened by more than 40% on average. In the domestic AI programming tool market, CodeBuddy relies on its three differentiated advantages of enterprise-level security compliance (MLPS 2.0 Level 3 certification), deep integration with the Tencent Cloud ecosystem, and a complete three-form product matrix of "plug-in + IDE + CLI" to occupy a solid market position among Tencent Cloud ecosystem users and high-compliance enterprises.

• 2. Company and developer information (1794 chars)

  Tencent CodeBuddy is independently developed by Tencent Cloud and is positioned as a core component of Tencent's "Industrial Internet" strategy in the field of software development. Wang Shengjie, head of AI products at Tencent Cloud Developers, previously stated in an interview with InfoQ that CodeBuddy’s core philosophy is to “be yourself, not someone else.” It emphasizes finding a differentiated path in the domestic AI programming tool market, rather than simply copying the functional models of overseas products.   From the perspective of development history, the evolution of CodeBuddy has gone through four main stages. Before 2018, it was an exploration period. The product relied on the IDE's own rules to provide similar function recommendations, and its capabilities were extremely limited. With the explosion of AI cloud services in 2022, Tencent launched the code completion capability, which is the true starting point of CodeBuddy. Between 2023 and 2024, the product will enter the agent stage, introducing the Agent architecture to support engineering understanding, knowledge base retrieval and inline dialogue, completing the transformation from a simple tool to an intelligent platform. From 2024 to 2025, CodeBuddy launched the Craft software development agent and began to support the concept of Vibe Coding (atmosphere programming). Users can describe their needs through natural language, and AI automatically completes multi-file code generation and rewriting.   2025 is a critical year for the comprehensive expansion of CodeBuddy’s product matrix. In July of the same year, CodeBuddy IDE started internal testing and became the first development workbench to realize the full-process AI integration of "product-design-R&D deployment". On August 21, CodeBuddy IDE took the lead in completing the access to the DeepSeek V3.1 model and launching public beta. On September 9 of the same year, Tencent Cloud released the new CLI tool CodeBuddy Code, making Tencent one of the few AI programming tool manufacturers in China and even the world to support three product forms at the same time.

• 3. Industry positioning (879 chars)

  CodeBuddy is in the fast-growing niche of AI programming tools. According to GitHub research data, AI-generated code already accounts for 41% of global code output, generating 256 billion lines of code in 2024. In the Chinese market, GitHub Copilot occupies about 64.5% of the market share, and Tongyi Lingma accounts for about 12.9% as a domestic alternative. IDC data shows that China's AI code generation market is ushering in an application explosion period in 2025, with more technology suppliers releasing and updating products based on the advantages of code engineering.   CodeBuddy's unique positioning in this market is as an "engineering agent platform" rather than a simple "code completion tool." Tencent Cloud emphasizes that in the domestic market, data security, code privacy and cloud compliance are the core value points that differentiate it from international competitors. At the same time, the product's compatibility with domestic mainstream IDEs and domestic code hosting platforms is also an important advantage.

• 4. Core functions (1594 chars)

  CodeBuddy provides four core functional modules, covering the complete development process from code writing to deployment and operation.   Code completion is the basic capability of CodeBuddy, which supports multiple modes such as generating code based on comments, completing inline code, completing the following based on the above, and completing code within function blocks. The technical dialogue function is based on situational awareness technology, providing reasoned answers and responses, supporting question recommendations and dialogue command operations, and can even directly generate code suggestions through dialogue. Craft's development agent is the differentiated core of the product, providing intelligent development assistance and supporting the full-process development model of "conversation as programming". Users only need to describe their requirements in natural language, and AI can automatically complete requirements disassembly, multi-file code generation and test integration.   In terms of engineering-level capabilities, CodeBuddy can understand and analyze multiple files in the project and provide developers with code suggestions that highly match the current project style. The unit test generation function supports mainstream frameworks such as pytest and unittest, and can automatically generate test cases. The intelligent review function assists in code review and automatically detects code problems. The code repair feature provides fix recommendations or automatic fixes for discovered issues. The one-click document generation function can automatically generate complete comments including parameter descriptions and return values ​​after selecting a function. The knowledge base Q&A function supports access to enterprise or personal knowledge bases and is compatible with RAG capabilities.

• 5. Three product forms (1455 chars)

  CodeBuddy is currently the only AI programming tool in China that supports three product forms at the same time. The three are targeted at different user groups and usage scenarios.   The IDE plug-in version is aimed at daily developers and can be installed as a plug-in into mainstream IDEs such as VS Code, Visual Studio, JetBrains, Cloud Studio, and WeChat Developer Tools. It provides real-time code completion, cross-file analysis and refactoring capabilities, and is embedded into existing development workflows. The independent IDE version (CodeBuddy IDE) is aimed at large teams and complex engineering projects. It focuses on engineering context understanding, task splitting and integration of production, design and research. It provides Plan Agent (intelligent demand planning), Design Agent (multi-modal design, supports direct conversion of Figma design drafts into code), Coding Agent (core code generation) and one-click preview functions. The CLI version (CodeBuddy Code) is oriented to automation and batch processing scenarios, can be integrated into CI/CD pipelines, and supports batch API replacement, full warehouse indexing, and Git operation automation.   CodeBuddy IDE's Figma design draft to code function is the core highlight of its front-end development capabilities. Users can directly import the Figma design draft into the IDE, and AI will automatically generate the corresponding maintainable HTML/React/Vue code, with a high degree of visual restoration. The IDE also integrates support for three mainstream UI component libraries: TDesign, MUI, and Shadcn, and provides a built-in browser preview function to view the effects in real time during the development process.

• 6. Technical architecture (1078 chars)

  CodeBuddy adopts a dual-model strategy, and the bottom layer is driven by Tencent Hunyuan T1 and DeepSeek R1 dual engines. Hunyuan T1 focuses on in-depth analysis and fast code suggestions, with speed as the priority; DeepSeek R1 focuses on comprehensive contextual search and bug diagnosis, with breadth as the priority. The international version additionally supports international mainstream models such as GPT and Gemini. In terms of context management, CodeBuddy adopts a hybrid strategy of "compression + external storage". Conversation and editing history are lightly compressed in local or temporary memory, similar to "streaming summary", retaining core task intentions and the latest code snippets.   CodeBuddy is also the first code assistant in China that supports MCP (Model Context Protocol), which can connect the entire end-to-end development process and support access to third-party MCP tools to expand the ecosystem. The Plan Mode function automates the entire process of requirement disassembly, code generation, and operation verification, and supports task-level checkpoint settings to ensure that a summary pops up for human confirmation before each step of the AI ​​operation, and the results are automatically verified before submission.

• 7. Performance (917 chars)

  According to Tencent Cloud's official data and technical evaluation report, CodeBuddy shows competitiveness on multiple key indicators. Coding efficiency is increased by about 40%, AI-generated code accounts for more than 40%, and R&D efficiency is improved by more than 16%. In complex task scenarios, China Merchants Bank used Craft mode to develop a travel assistant APP, which shortened the time from 2 days to 2 hours.   Compared with domestic competing products, CodeBuddy has obvious advantages in security compliance. The SQL injection detection accuracy reaches 99.2%, the false positive rate is reduced by 67%, and it has passed the Level 3 Certification of MSW 2.0. It is the first choice for high-compliance scenarios such as finance and government affairs. Response latency is about 1.5 seconds, slightly higher than ByteDance Trae’s 0.8 seconds, but security and compliance are stronger. The cost of private deployment is 150,000 to 800,000 per year, which is lower than Tongyi Lingma's 200,000 to 1 million per year, and Tongyi Lingma does not support private deployment.

• 8. User experience (1083 chars)

  According to multiple developer test reports, CodeBuddy’s experience evaluations vary in different development fields. The front-end development aspect has the highest score, about 9/10. The UI design level is high, the code quality is good, and the functions are complete. In particular, the Design to Code ability has received unanimous praise. The score for requirements analysis is about 8/10. The PRD document structure is complete and the functional modules are clearly disassembled. The back-end development score is about 7/10. The database design is reasonable and the deployment configuration is considerate, but the ability to process complex business logic is limited.   In actual testing, developers found that the generated code often did not work out of the box and required debugging. There is a lack of production-level error handling and security considerations, and there are basically no performance optimization suggestions. Error location is not accurate enough, the quality of repair suggestions is unstable, and multiple rounds of dialogue are often required to obtain available repair solutions. "Integration of production, equipment and research" still feels fragmented in actual use, and the connection between the modules is not smooth enough.

• 9. Version division (839 chars)

  CodeBuddy uses a layered business model and is divided into four editions.   The personal version is for individual developers and uses Tencent Cloud shared VPC. Its core benefits include real-time code continuation, annotation code generation, intelligent technical dialogue, unit test generation and knowledge base Q&A. The personal version is currently free for a limited time. The Ultimate Edition is for enterprises or teams, and also uses Tencent Cloud Shared VPC. It includes all the functions of the Personal Edition, and adds member authorization management and enterprise research performance dashboards. The exclusive version is aimed at enterprises with high security requirements. It uses Tencent Cloud's exclusive VPC and includes all the functions of the flagship version, adding exclusive network access and exclusive enterprise plug-ins. Private subscriptions offer fully on-premises deployments that meet the highest levels of data security and compliance needs.

• 10. Points billing system (664 chars)

  CodeBuddy adopts a resource quota management mechanism based on credits. Credits are used to measure resource consumption during the execution of AI tasks. Different models and task complexity will affect the consumption of points. Advanced models consume more points when processing the same number of Tokens, simple tasks consume less, and complex tasks such as long code analysis and multi-round dialogues consume more.   For point deduction of the personal version, priority will be given to the points quota of the personal trial version. When insufficient, the resource package that expires first will be deducted first. The points that expire first will be used first for the flagship and exclusive versions, including monthly fixed usage and enterprise volume-added packages.

• 11. positive review (1452 chars)

  Feedback from developers with different experience levels shows obvious polarization characteristics, but positive comments dominate.   For novice developers, CodeBuddy is evaluated as an "ideal development aid" that can help easily understand the code structure and better master programming languages ​​​​such as Go and Python. For senior developers, CodeBuddy can provide corresponding suggestions based on the coding style of the engineering project. These suggestions are highly consistent with the current project and can automatically generate interfaces through simple comments, "greatly improving programming efficiency." Comments from front-end experts include "shortening coding time", "quickly analyzing and solving problems", "improving code quality", "lowering the threshold for learning new syntax" and "accurate recommendations are very easy to use".   In terms of enterprise-level applications, usage data from Tencent Healthcare Team shows that the weekly code completion rate reaches 39.81%, the weekly adoption rate reaches 31.63%, the weekly activity rate reaches 96.82%, and nearly 40% of the code is written by CodeBuddy. Well-known companies such as Xiaomi Group, Midea, Honor, and Xiaoetong are also using CodeBuddy. The case of China Merchants Bank shows that using Craft mode to develop a travel assistant APP took 2 days to 2 hours.   Test reports from the developer community generally believe that although CodeBuddy still has room for improvement, it has demonstrated stronger front-end capabilities and more complete full-stack development support than competing products. It is currently one of the domestic AI programming tools closest to the "usable" standard.

• 12. negative feedback (1127 chars)

  Usability issues are the most negative factor reported by users. The memory usage soared during plug-in initialization, a chat window popped up by default every time the project was opened, and prompt messages during auto-completion blocked normal code jumps. These problems affected the developer's workflow experience. Some users have reported that the generated code will insert irrelevant strings, such as "Jisu E-sports" and "Speed ​​Racing Lottery Live Broadcast" and other irrelevant content mixed into the DeepSeek V3.1 model training data, which has a certain impact on user trust.   At the functional level, the generated code cannot be fully used out of the box and requires developers to debug it themselves. There is a lack of production-level error handling and security considerations, limited complex business logic processing capabilities, and basically no performance optimization suggestions. Error location is not accurate enough, the quality of repair suggestions is unstable, and multiple rounds of dialogue are often required. It can be seen from the plug-in update log that early versions had problems such as high memory usage, unstable model switching, and LSP shutdown due to insufficient resources. Although improvements have been made in subsequent versions, there is still room for optimization.

• 13. Industry status (496 chars)

  In international independent reviews, CodeBuddy scores above average but lags behind the first tier. In aitrendblend.com's 2026 comprehensive review, CodeBuddy received a score of 7.8/10, ranking seventh among 7 Chinese AI programming tools, behind DeepSeek V3 (9.4 points), Kimi 2.5 (9.1 points) and Qwen2.5-Coder (8.9 points). Reviews pointed out that CodeBuddy's context window is 32K, which is significantly lower than the 128K of competing products, limiting its ability to handle large code bases, and the overall coding quality lags significantly behind the first echelon.

• 14. media opinion (1004 chars)

  Coverage of CodeBuddy in mainstream Chinese tech media has been generally positive. InfoQ pointed out in an in-depth interview that CodeBuddy has become an important alternative to domestic products by virtue of its dual-model strategy and engineering agent positioning in the context of international tools such as Claude "blocking" the Chinese market. 36Kr reported on the internal beta launch of CodeBuddy IDE, emphasizing its core selling point of being “the first to realize full-process AI integration of product-design-R&D deployment.” Geek Park reported on the release of CodeBuddy Code CLI, commenting that Tencent became “the first manufacturer to support three forms of AI programming.”   Evaluation articles in the developer community generally believe that CodeBuddy has differentiated advantages in terms of Design to Code capabilities in the front-end development field and Tencent Cloud ecological integration, but there is still room for improvement in back-end quality and complex project processing capabilities. Compared with ByteDance Trae, CodeBuddy is more mature in front-end UI generation, and its overall usability is closer to practical standards.

• 15. Competitive product landscape (725 chars)

  In China's AI programming tool market, CodeBuddy's main competitors include Tongyi Lingma (Alibaba Cloud), Trae (ByteDance), Wenxin Kuaima (Baidu) and CodeGeeX (Zhipu AI).   The advantage of Tongyi Lingma lies in Alibaba ecological integration and multi-file batch modification capabilities. It is suitable for enterprise-level deployment of large enterprises, but the cost structure is opaque. Trae is known for its multi-model switching and algorithm accuracy (92%). Its free strategy is attractive to individual developers, but it lacks commercial support. CodeBuddy's differentiated advantages lie in enterprise-level security (Class 2.0 Level 3 certification), deep integration with Tencent Cloud ecosystem and a complete three-form product matrix, which is suitable for Tencent Cloud ecosystem users and high-compliance enterprises.

• 16. Tencent internal application data (342 chars)

  According to the "2025 Tencent R&D Big Data Report", more than 90% of Tencent's engineers are using the AI ​​programming assistant CodeBuddy to assist in writing code. 50% of new codes are generated with AI assistance, and the overall coding time is shortened by more than 40%. This internal large-scale application experience provides rich real-life scenario feedback for continuous product iteration.

• 17. Performance and stability controversy (316 chars)

  As a rapidly iterating product, CodeBuddy still faces challenges in terms of user experience maturity. Several reviewers pointed out that CodeBuddy's promised time-saving tool features may instead cost developers more energy due to problematic code or frequent crashes. This "gap between promise and reality" is one of the main points of controversy faced by the product.

• 18. Data quality issues (311 chars)

  Users discovered that when using the DeepSeek V3.1 model, the code generated in UI development would insert irrelevant strings. This problem was caused by insufficient cleaning of the model training data. Although Tencent has dispersed risks to a certain extent through its dual-model strategy, data quality issues still affect users' trust in product reliability.

• 19. Context window limits (279 chars)

  Compared to competing products, which generally reach 128K context windows, CodeBuddy’s 32K context limit significantly limits its ability to handle large code bases. In scenarios where cross-file reconstruction or full warehouse code analysis is required, this limitation may affect the user experience and output quality.

• 20. Ecological lock-in risk (219 chars)

  CodeBuddy's core advantages are highly dependent on the Tencent Cloud ecosystem, and its practicality outside the Tencent Cloud ecosystem may be greatly reduced. For non-Tencent Cloud users, the differentiated value of this product is relatively limited.

• 21. Recommended usage scenarios (874 chars)

  Tencent Cloud ecosystem users are the best applicable group for CodeBuddy. The product is seamlessly integrated with Tencent Cloud services such as CloudBase, EdgeOne Pages, Cloud Functions, and Cloud Database, which can significantly improve development efficiency. High-compliance companies are also suitable to choose CodeBuddy. Its Class III certification and privatized deployment solutions can meet the compliance requirements of highly regulated industries such as finance, government affairs, and medical care. For start-up teams with rapid MVP verification needs, CodeBuddy’s Craft mode can shorten the application development cycle from days to hours. Novice developers can use CodeBuddy's conversational programming and code explanation capabilities to accelerate the learning curve and lower the barrier to entry to programming. Users with non-technical background can take advantage of the natural language programming capabilities of CodeBuddy IDE to achieve basic application development needs.

• 22. Scenarios that need to be used with caution (429 chars)

  Development of critical production systems is not suitable for relying on CodeBuddy, and the debugging costs and potential risks of generated code need to be carefully evaluated. Complex business logic has limited processing capabilities and may require developers to make a lot of corrections and improvements. It is recommended to adopt a more conservative AI-assisted strategy for core code development of large full-stack teams, and CodeBuddy can be used as an auxiliary tool rather than the main tool.

• 23. alternative (272 chars)

  For general advanced coding tasks, DeepSeek V3 and Qwen2.5-Coder outperform CodeBuddy in terms of coding quality. For scenarios that require processing large code bases, competing products with 128K context are more suitable. For non-Tencent Cloud ecosystem users, Tongyi Lingma or Baidu Comate may be better choices.

• 24. Summarize (1291 chars)

  As an important player in the domestic AI programming tool market, Tencent CodeBuddy relies on its three differentiated advantages of dual-engine-driven technical architecture, "plug-in + IDE + CLI" three-form product matrix, and deep integration of the Tencent Cloud ecosystem, occupying a solid position in the Tencent Cloud ecosystem user and high-compliance enterprise market segments. More than 90% internal usage and a million-level total user scale verify the basic usability of the product, while the Class III certification 2.0 and 99.2% SQL injection detection accuracy reflect its core competitiveness in the security compliance dimension. However, the 32K context window limit, the user experience maturity issues caused by the rapid iteration stage, and the reality that generated code requires secondary debugging, indicate that CodeBuddy is still in the process of climbing from "available functions" to "excellent experience."   Looking to the future, as competition in the domestic AI programming market intensifies, CodeBuddy's growth space depends on whether it can achieve breakthroughs in three dimensions: first, continue to improve the quality stability and out-of-box ratio of code generation, and reduce users' debugging costs; second, make up for the shortcomings of context windows and strengthen the processing capabilities of complex projects; third, deepen the applicability outside the Tencent Cloud ecosystem and explore a wider range of developer groups and market scenarios.

No in-depth report yet

• 

  Kathryn_Phillips_Max4

  —

  5分钟搞定一个todo清单，这效率有点离谱了吧

---

• 

  MWhite_995

  —

  Craft模式确实强，但每个月这点免费次数是真的难受，强烈建议充个会员

---

• 

  KoalaKrypto643

  —

  免费额度才50次/月，根本不够用啊，连两天都撑不过

---

• 

  Stephanie_Morgan_Pro

  —

  生成的效果第一眼看着有点简陋，太简约了，不是那种精致的风格

---

• 

  Jonathan_PatelIII

  —

  对比了下Cursor，感觉CodeBuddy在前端UI这块更强，设计稿直接转代码太香了

---

• 

  SPeterson5

  —

  作为一个干了10年的后端开发，CodeBuddy给我的感觉是这回腾讯真的没忽悠人，确实挺能打。前端开发这块UI生成能力是真的强，评分我给9/10

---

• 

  Paul.James

  —

  腾讯云生态用户首选不是吹的，CloudBase一键部署香得不行

---

• 

  徐桂

  —

  之前用Copilot，现在换CodeBuddy了，响应速度是真的快

---

• 

  Katherine_Kim_Max

  —

  用Craft模式做了个小程序，本来要搞两天的活，两小时就搞定了，真香

---

• 

  蝴蝶_2

  —

  吹爆Figma设计稿转代码这个功能，太好用了，设计师和前端都解放了

---

• 

  LanceMoore

  —

  等保2.0三级认证这个是真的香，金融行业的懂的都懂

---

• 

  Diane.Anderson_7

  —

  有点过头了没想到腾讯能做成这样，整体体验超出预期，好的有点离谱。每次技术浪潮到来都会有人恐慌，但CodeBuddy证明了AI编程确实在改变开发方式

---

• 

  Kelly_MorrisIII302

  —

  说实话比字节Trae好用，至少前端这块碾压

---

• 

  greenladybug748

  —

  生成代码不能开箱即用，总要调试下，这点有点烦

---

• 

  TerryHughesJr

  —

  复杂业务逻辑还是差点意思，得自己改

---

• 

  Walle_tWolf234

  —

  微信开发者工具直接集成这个确实牛，腾讯的生态优势还是明显

---

• 

  KyleBell007

  —

  内存占用有点高，每次打开项目都卡一下

---

• 

  Heather_YoungK

  —

  个人版限时免费，但额度太少了希望多给点

---

• 

  AWalkerII0

  —

  用了三个月了，CodeBuddy已经成为我日常离不开的工具。中小型项目用它来快速验证想法真的很方便，PRD写完直接出代码，测试覆盖率也能达到95%以上，说实话有点后悔没有早点用

---

• 

  Lauren_Gonzales_77

  —

  新手友好度拉满，教我表弟用这个学Python，两周就会写猜数字游戏了

---

• 

  Jeffrey_Foster

  —

  错误定位有时候不太准，修复建议质量不稳定，得来回改好几轮

---

• 

  goldencat870

  —

  产设研一体化听起来很美，但实际用起来还是有点割裂感

---

• 

  Andrea.SimmonsSr

  —

  国内首选不是吹的，编码效率提升40%不是盖的

---

• 

  MichellePrice

  —

  回不去了，Copilot已经吃灰两个月了

---

• 

  Donald_Hart520

  —

  SQL注入检测准确率99.2%这个数据是认真的吗，实测了一下好像确实挺准

---

• 

  Katherine.Wright_2022

  —

  性能优化建议基本没有，这点差评

---

• 

  Madison_CastilloIII

  —

  腾讯内部超过90%工程师在用，这个数据应该是真的，我们team现在全员都在用

---

• 

  DeltaDefi386

  —

  Prompt优化这个功能是真的好用，先让AI优化需求再编码，输出质量提升明显

---

• 

  Arthur.AndersonSr

  —

  说实话，对比DeepSeek V3这些第一梯队，CodeBuddy的编码质量还是差点意思，上下文窗口32K也偏小

---

• 

  Barbara.RodriguezIII

  —

  非技术背景也能用这个开发应用了，我行政部的同事用它做了个内部预约系统，腾讯这波是真的在推进人人都是开发者的愿景

No reviews yet