In VSCode and JetBrains, users of Codeium are able to activate Codeium's Indexing Engine—which generates embeddings for the user's local codebase that are stored on a local index. The Codeium extension can then query this index via a vector / similarity search in order to retrieve the most relevant context items to provide better context-aware Autocomplete and Chat.
However, this system comes with some caveats: while the embedding is done remotely, the process of iterating through the files in the codebase to create the index is performed locally on the user's machine. For large repositories this process can consume a lot of RAM on consumer hardware and querying the index for large repositories can take a while. Furthermore, if the user depends on or wants to ask about a repository that they do not have locally cloned, the autocomplete and chat models will be unaware of it as the code is not present in the local index.
Organizations on Codeium Teams can use Codeium's Indexing Service to globally import all the relevant repositories. The indexing and embedding is then performed by Codeium's servers (on an isolated tenant), and once the index is created, it is available to be queried by any member of the Team.
Codeium Indexing is currently in private-beta. Please contact Codeium if you'd like to enable it for your team.
Adding a Repository
From https://codeium.com/indexing you can add a repository to index. Currently we support Git repositories from GitHub, GitLab, and BitBucket.
You can choose to index a particular branch and to automatically re-index the repository after some number of days.
We clone the repository in order to create the index, but crucially once we finish creating embeddings for the codebase we delete all the code and code snippets. We don't persist anything other than the embeddings themselves, from which you cannot derive the original code.
Furthermore, all indexing and embedding is performed on a single-tenant instance—nothing about the indexing process is shared between multiple Codeium Teams customers.