Integrated RAG / Indexing / vectors like Cursor Indexing or Continue

[giggoretta]

Hi Roo Code team,

I’d like to request a feature to enhance Roo Code with vector-based capabilities, such as RAG, indexing, or vector search, similar to what Voyage Code 3 with Qdrant, Cursor Indexing, or Continue offers. This would allow for efficient semantic search and context retrieval of code snippets within the codebase, making it faster and more accurate to handle complex tasks. For example, we could provide .txt files (code snippets, manuals, etc.), and Roo Code, at a rate of X$/1M tokens, would use the most advanced embedding AI to convert the project, manuals, code snippets, and even the prompt itself into a vectorized format in a 1024 or 2048 dimensional space, utilizing the maximum float value of 3.4028235e+38 for precision.

To take this to an even higher level, an integrated AI Agent could be added, which, using the most advanced AI models, would vector-specifically tag and comment on ingenious, logically, or mathematically valuable code that might be poorly documented or not commented at all. This would make it much easier to work with niche (lesser-known) scripting languages as well. I believe this addition would make Roo Code even more powerful for developers working on intricate coding workflows. Thanks for considering this!

[BradKML]

I think this is a beta-test feature for RooCode, no?

[BradKML]

Also two cross-refs https://github.com/FareedKhan-dev/all-rag-techniques #3979

[ChuKhaLi]

Roo has codebase indexing as experimental feature from version 3.18.3
https://github.com/RooCodeInc/Roo-Code/releases/tag/v3.18.3
https://docs.roocode.com/features/experimental/codebase-indexing

[BradKML]

Best to keep this on so people can experiment on diverse set of RAGs to find the best options

[giggoretta]

Roo has codebase indexing as experimental feature from version 3.18.3 https://github.com/RooCodeInc/Roo-Code/releases/tag/v3.18.3 https://docs.roocode.com/features/experimental/codebase-indexing

This is fantastic news! It means we can expect the following features later in Roo Code:

==> Context-based code completion (context derived from the project’s own codebase and other vectors, such as syntax manuals, other projects, and syntax-specific coding tricks).

==> Vector-based code generation from the chat window, utilizing the project’s context and fine-tuning-like files (especially helpful for lesser-known coding languages, where the large OpenRouter models have insufficient training data).

==> Project-based intelligent code search, quickly finding relevant code snippets using vector-embedded content.

==> Generating codebase-specific recommendations considering the project’s unique coding style, patterns, and requirements.

==> Intelligent debugging assistance based on vector analysis and previously encountered similar errors.