Baidu Comate opens up the plug-in ecosystem, and the era of intelligent code assistant customization is coming

On March 1, Baidu Comate, Baidu’s intelligent code assistant, added two major capabilities: “Comate” open platform and AutoWork “private R&D assistant”, becoming the first free trial in the industry. With this release, Baidu Comate will be more in line with the software research and development site, and will better meet the needs of enterprise customized development through an easy-to-use R&D platform, rich plug-in basic capabilities, independent customization capabilities, and enterprise access to private domain knowledge and own capabilities. needs, helping companies create intelligent code assistants suitable for their own companies at low cost, greatly improving software development experience and efficiency.

"Comate" open platform realizes the in-depth integration of enterprise private domain knowledge, third-party capabilities and programming on-site, directly accessing the first work site of R&D personnel, and quickly applying large model capabilities without repeated construction, unlimited Expand platform scenarios to make the intelligent code assistant more adaptable to the company's business knowledge, team specifications, standard processes, and R&D management.

Through the private domain knowledge enhancement function of the "Comate" open platform, it is possible to connect enterprise private domain knowledge. If an enterprise has specific code specifications or writing methods, this knowledge can be uploaded to the Comate open platform and bound with relevant functions. Comate can make full use of the enterprise's private domain knowledge and further improve its accuracy. It is reported that, judging from Baidu’s internal business usage, after applying this function, the recommendation accuracy rate generally increases by more than 10%. Among them, the accuracy rate of PaddlePaddle code generation increased from 60.5% to 81.5%, fully demonstrating Comate's powerful ability in private domain knowledge enhancement.

Through the "Comate" open platform, you can also directly connect to third-party services, capabilities, etc., directly connect all aspects of software development and auxiliary capabilities/knowledge to the programming site, and create "own" tailored for the organization The intelligent code assistant is currently connected to GitHub, GitLab, Gitee and other plug-ins to help users complete operations intelligently.

Through the immersive use of these plug-ins in Comate, the programming experience and efficiency can be significantly improved. Taking access to GitHub as an example, users can use the GitHub plug-in to automatically obtain the changes in the current project, automatically analyze the issue list of the GitHub warehouse, and generate submission messages that match needs.

Intelligent analysis of code changes

Associated requirements, learning history submission information format

This time Another major upgrade capability is AutoWork. Following its appearance at the Wave Summit 2023 Deep Learning Developer Conference, after 2 months of iteration, AutoWork has been opened to enterprise SaaS users, further strengthening its characteristics as a "private R&D assistant". It can deeply understand the local code base and private domain knowledge within the organization. Developers only need to clarify the development "goals" and "intentions", and AutoWork can automatically retrieve the necessary background knowledge, independently analyze product requirements, match the best solution and generate code to better complete development instructions. Among them, the two functions of "intelligent question and answer" and "generating code based on private domain knowledge" are very prominent.

The "Intelligent Q&A" function mainly allows developers to "chat" directly with the code base. Faced with the vast code base, whether it is to understand the overall architecture, explore specific implementation logic, or sort out the complex dependencies between codes , AutoWork can give clear and accurate feedback in real time, allowing developers to quickly grasp the full picture of the project. For example, when a new code base needs to be mastered, developers do not need to read through a large amount of documents to understand complex logic. They only need to directly ask AutoWork to sort out the code structure, and a clear and complete explanation will be given, realizing "ask where you don't understand".

#In addition, every organization relies on a large amount of "private knowledge" to achieve business success. However, in past human-machine collaborations, machines had “little knowledge” of private domain knowledge and were unable to apply it. This means that on the one hand, humans still bear a huge amount of labor in transforming private domain knowledge, such as learning, understanding, association, application, etc.; on the other hand, it also indicates the huge potential of intelligent assisted knowledge applications. Based on this, this upgrade of AutoWork focuses on seamlessly connecting the organization's private domain knowledge to the code assistant, that is, by mounting specified development documents, requirements documents, API documents, etc., and applying the capabilities of the Wenxin large model, these private knowledge Quickly understand, analyze and apply domain R&D knowledge to achieve automatic generation from requirements to code. As a developer, you no longer need to spend a lot of time reading documents to achieve "one-click" efficient programming.

For example, by mounting Wenxinyiyan’s development documents, the code generation for calling Wenxinyiyan’s API can be quickly completed. This not only reduces the time developers need to manually read and understand a large number of documents, but also greatly reduces the time required to manually read and understand a large number of documents. Understand deviations leading to errors and rework.

AutoWork combines private domain knowledge to generate code

In this upgrade, AutoWork de-emphasizes grammar learning and direct configuration files in code generation, single test design and generation, etc. In terms of multiple basic experiences, it also further enhances human-machine collaboration capabilities, greatly improving developer experience and efficiency.

Baidu Comate continues to deepen its attention to the needs of enterprise customers. On the basis of its existing advantages such as data security, rapid interaction, and flexible deployment, it has also realized the new capabilities through the two latest launches. Enterprise private domain knowledge connection and in-depth satisfaction of personalized needs.

To learn about Baidu Comate intelligent coding, visit the official website https://www.php.cn/link/705a254246ff6850de83ad6a8e461c0a to get it.

The above is the detailed content of Baidu Comate opens up the plug-in ecosystem, and the era of intelligent code assistant customization is coming. For more information, please follow other related articles on the PHP Chinese website!