The VectorCode extension just got better!

[Davidyz]

tl;dr: with the 0.7.0 release of VectorCode, you can now let the LLM handle the vectorisation! This makes using VectorCode with codecompanion easier than ever, because you no longer have to manually create the embeddings from the CLI. The following is an example of using CodeCompanion and VectorCode to explore the awsome fzf-lua repository:

Feel free to visit the wiki to learn more about the updated VectorCode extension, and I'd love to hear your feedback on this!

[daviareias]

Hello I'm trying to understand how to use workspace in codecompanion, and I'm not sure what is vectorcode or how it helps.

From the video it seems like you sent some prompt and it generated for you? The main reason I'm trying to learn workspaces is because I want to have control over what I send outside my machine.

[Davidyz]

Hi, VectorCode is a RAG utility that nicely integrates with CodeCompanion to provide knowledge about the project you're working on to the LLM. It allows the LLM to perform semantic searches on your local repositories and returns contexts that help the LLM understand the project, so that it won't hallucinate as much and will provide better solutions for your coding tasks. The video shows how I asked the LLM to create a database for fzf-lua and search for dap-related source code to provide project-specific and up-to-date answers.

In terms of privacy control, this is really where Vectorcode shines because you can make it run completely locally, and you have 100% control over what code to add to the database (by using the vectorcode vectorise command in CLI or the @vectorcode_vectorise tool like the demo). There are also other fancy features like result summarisation, which allows you to use a local LLM (or hosted LLM, depending on your config) to rewrite the retrieval results so that your main chat LLM will only see the summaries, not the actual code.