Is there an open source version of Cursor from previous generations?

[tahababou12]

There used to be an open-sourced version of Cursor a few months ago, but I cannot find the repository anymore. I looked back in this repository's commit history and managed to clone it back to how it was before.

Unfortunately, after trying to run ./setup.sh - it appears that the following links are now invalid:

wget https://cursor.so/resources.zip
wget https://cursor.so/lsp.zip

If you could share Cursor's previous generation open-source repo - I'd greatly appreciate it!

[ymd65536]

I was looking for the zip file that gets to setup.sh as well.

[danperks]

Unfortunately, that version is no longer support.

It relied on a backend server which is no longer functional and has had significant upgrades to the latest Cursor version.

Is there something you were looking to implement or do in the old Cursor, that the newest version doesn't support?

[tahababou12]

Thank you for the information. I am particularly interested in understanding how Cursor's backend architecture ingests GitHub repositories into vector databases to support a Retrieval-Augmented Generation (RAG) framework. My goal is to explore how this process indexes codebases and enables efficient query handling for more relevant and context-aware results when working with large repositories.

I am currently building a RAG chatbot and would like to integrate Cursor as a third-party API. Understanding how the previous version of Cursor handled repository ingestion would help me greatly in achieving that.

Does the current version of Cursor offer any documentation or insights on this aspect, or has the architecture changed significantly in the latest release?

[danperks]

As a closed source project, I don't believe there is too much information regarding how the backend of Cursor is implemented.
That's kind of the "secret sauce" that makes Cursor so powerful.

One of the devs may be able to comment more, but there's no public API available to query for any integrations. You may find some clues in the previous commits in this repository, but I think that's as far is public for now, sorry!

[tahababou12]

I see, thank you, Dan! I suppose one of the key ingredients in the 'secret sauce' lies in how the code files are chunked and embedded. I'll look into the commit history for any hints and continue exploring approaches for code ingestion and indexing. If there's any additional insight on embedding strategies or if the team ever decides to offer a public API for code-related integrations, I'd be very interested in hearing about it!

Thanks again for your help.