The AI Chat model has been upgraded to Qwen2.5 Coder, the input interaction has been upgraded and now supports multiple context types, and a new multimodal image chat mode has been added.
, making a little more progress every day.
AI Chat has been upgraded to the latest Qwen2.5 Coder model, delivering significant improvements in programming performance and efficiency. Its flagship code model achieved the best open-source results across more than ten benchmark evaluations, making it the world's most powerful open-source code model. Generation speed has more than doubled compared to before. Update to the latest plugin to try it out.
The input interaction in the AI Chat window has been comprehensively upgraded to support adding contexts independently, allowing users to freely combine natural language requirement descriptions with contexts. In addition, supported context types have been expanded to include #file, #codeChanges, #gitCommit, #teamDocs, and #image. Developers can flexibly choose one or more files to ask questions, or ask about and review current code changes and commits in the project, significantly improving problem-solving efficiency and accuracy.
The new multimodal chat mode is now live. It can analyze image content or generate code suggestions based on images. You can add images as context by selecting, dragging, or pasting images and then asking questions. For example, if you encounter an error, you can paste a screenshot directly into the input box to analyze the issue and receive a solution. You can also add design mockups or web page screenshots and enter generation requirements to receive suggested code.
AI Chat retains conversation memory. If you feel a particular question is not ideal and are concerned about its impact on the overall session, you can directly delete that question and its reply. Likewise, if you want to share a reply with someone else or save it, you can click to copy it directly.
When creating enterprise custom instructions, we have added support for #file, #codeChanges, and #gitCommit as context for instruction assembly. When designing instructions for scenarios such as asking questions about code files or performing code reviews, enterprise administrators can freely combine these contexts to achieve more precise and efficient instruction configuration, improving instruction flexibility and practicality for various scenarios.
To better meet enterprise needs in code generation enhancement scenarios, we have added the ability to import repositories via Git URL. This feature enables users to quickly and easily connect existing repositories to our platform, simplifying the repository data import process. Users only need to provide the Git repository URL to seamlessly connect the repository, greatly improving data management efficiency for enterprise code generation enhancement scenarios.
In Enterprise Dedicated Edition, the knowledge management module now supports enterprise-owned on-cloud or on-premises storage and vector databases. Enterprise administrators can now manage data and resources more flexibly, which not only improves controllability and security of data management, but also enhances the flexibility of enterprise data and resource management.
1. AI Chat model upgraded to the latest Qwen2.5 Coder
AI Chat has been upgraded to the latest Qwen2.5 Coder model, delivering significant improvements in programming performance and efficiency. Its flagship code model achieved the best open-source results across more than ten benchmark evaluations, making it the world's most powerful open-source code model. Generation speed has more than doubled compared to before. Update to the latest plugin to try it out.
2. Input interaction upgraded, supporting multiple context combinations
The input interaction in the AI Chat window has been comprehensively upgraded to support adding contexts independently, allowing users to freely combine natural language requirement descriptions with contexts. In addition, supported context types have been expanded to include #file, #codeChanges, #gitCommit, #teamDocs, and #image. Developers can flexibly choose one or more files to ask questions, or ask about and review current code changes and commits in the project, significantly improving problem-solving efficiency and accuracy.
3. Added multimodal image chat mode
The new multimodal chat mode is now live. It can analyze image content or generate code suggestions based on images. You can add images as context by selecting, dragging, or pasting images and then asking questions. For example, if you encounter an error, you can paste a screenshot directly into the input box to analyze the issue and receive a solution. You can also add design mockups or web page screenshots and enter generation requirements to receive suggested code.
4. AI Chat supports conversation deletion and copying
AI Chat retains conversation memory. If you feel a particular question is not ideal and are concerned about its impact on the overall session, you can directly delete that question and its reply. Likewise, if you want to share a reply with someone else or save it, you can click to copy it directly.
5. Enterprise custom instructions now support multiple context types
When creating enterprise custom instructions, we have added support for #file, #codeChanges, and #gitCommit as context for instruction assembly. When designing instructions for scenarios such as asking questions about code files or performing code reviews, enterprise administrators can freely combine these contexts to achieve more precise and efficient instruction configuration, improving instruction flexibility and practicality for various scenarios.
6. Enterprise knowledge management now supports Git URL repository import
To better meet enterprise needs in code generation enhancement scenarios, we have added the ability to import repositories via Git URL. This feature enables users to quickly and easily connect existing repositories to our platform, simplifying the repository data import process. Users only need to provide the Git repository URL to seamlessly connect the repository, greatly improving data management efficiency for enterprise code generation enhancement scenarios.
7. Enterprise Dedicated Edition knowledge management supports self-owned storage and data
In Enterprise Dedicated Edition, the knowledge management module now supports enterprise-owned on-cloud or on-premises storage and vector databases. Enterprise administrators can now manage data and resources more flexibly, which not only improves controllability and security of data management, but also enhances the flexibility of enterprise data and resource management.
8. Other experience improvements
- Unit test generation on the plugin side now supports awareness of repository POM file configuration, improving unit test generation effectiveness
- Simplified AI Chat shortcut instruction names on the plugin side for easier feature recognition
- The Enterprise Edition knowledge upload OpenAPI supports batch upload
- The Enterprise Edition department/user data query API now includes chat code adoption-related data