The Langchain indexing API relies on having the ID computed during the hashing step be saved into the database so it can be used compare/add/remove records as needed.

The AzureCosmosDBNoSqlVectorSearch is currently ignoring any IDs coming in from the data, and instead computing its own ID using UUID:

This results in errors when using the Indexing API.

I provided a PR fix for this that we're using in the langchain-community repo: langchain-ai/langchain-community#62. Happy to port this to this new instance if that's welcome.