Dust Community Icon
Join the Community
Help

Best Practices for Code Uploads in Dust: Seeking Community Insights

Avatar of Kheira BenmeridjaKheira Benmeridja
·
·

Hi Dust Community! 👋 I'm working on a documentation assistant project and looking for best practices regarding code upload. Here's my specific use case: Current Setup: - Using Dust to analyze UI screenshots and data models - Need to give the assistant access to our GitLab codebase for: * Business rules extraction * Workflow analysis * Field validation rules Current Plan:As I don't want to make a connection via API, I'm planning to use Dust Folders to upload our code files Questions for the Community: 1. Has anyone implemented a similar setup for code analysis? 2. Any tips for organizing code files in Dust Folders for optimal assistant understanding? 3. What's your experience with large codebases in Dust? Any specific limitations or workarounds you've found useful? 4. Any recommendations for structuring the files to help the assistant better understand code relationships? Any insights, examples, or lessons learned would be greatly appreciated! 🙏

1 comment

  • Avatar of Remi
    Remi
    ·
    ·

    Hello Kheira Benmeridja 👋 Looks like an awesome exploration ! So far, I only worked with the native Github connection. While it allowed us to to some interesting things and help in many use cases, it wasn't a replacement for Github Copilot (more a complementary). You can read more in this article: https://blog.dust.tt/integrating-ai-workflows-alan/

    4. Any recommendations for structuring the files to help the assistant better understand code relationships?

    This is typically where Dust isn't perfect. If you use the "search" tool. It will do vector search on all of the files and for the most part not take into account the file structure. It can still be surprisingly good in many use cases but it's definitely still a limitation. Let us know how it went in 🔒[private feed]! 🤩 Let me know if you have any other questions!