Does using an OpenAI integration send your data to OpenAI

[mostafaibrahim17]

Hello, a beginner here. I am trying to understand if using an OpenAI integration sends your "training data" to OpenAI or not. To give some context, I am using a FAISS memory store that fits on 25,000 pages of my private data, and I am using an OpenAI integration for queries. I understand that the query itself is of course sent to OpenAI's API, but do parts of my own private data get also sent?

[oddrationale]

In order to create your vectorstore, you need to create embeddings of your 25,000 pages of documents. If you are using OpenAI to create the embeddings, then you are indeed sending the text of the 25,000 documents to the OpenAI Embeddings API and receiving the embeddings back to store in your FAISS vectorstore.

In addition, the queries/prompts you send to OpenAI will typically contain the relevant excerpts from your 25,000 documents as part of the context of the query. So those excerpts are also being sent to OpenAI.

Hope that helps!

[skcoirz]

+1 to what @oddrationale mentioned above. For embedding build up, alternatively, you can use a local model, instead, for privacy advantage. HuggingFace hosts a ton of embedding models. (link: https://huggingface.co/models?sort=downloads&search=embedding)

[dannyb2100]

I thought I would add to this conversation, instead of starting a new one.

We want to use LangChain with an LLM, ideally OpenAI or AzureOpenAI.

The aim is to use the SQLAgent to interact with the database.

Are there any charts, diagrams or documentation on where data flows? Understanding the full interaction and how things are processed fully would be nice.

Using this as an example: https://python.langchain.com/docs/integrations/toolkits/sql_database
E,g a user enters a prompt requesting "What month had the most sales in 2022".

Where the agent is coming up with thoughts and observations, what data is sent to the API and what is done locally? Is the schema sent to the API? Parts of it?