PANDAS API KEY needed (and used!!!) if agent.train is utilized

[flashtheman]

System Info

pandasai: v2.0.33
azure openai with gpt-4, api version 2024-02-01

🐛 Describe the bug

I defined the pandas ai api key like this, because it seems there is a bug that requires it in combination with azure open ai api (I get "pandasai.exceptions.MissingVectorStoreError: No vector store provided. Please provide a vector store to train the agent". otherwise):
os.environ["PANDASAI_API_KEY"] = "xxx"

I refer to the llm, which is azure open ai:

agent = Agent(df, config={"llm": llm})

When I train, it writes back my training data to the pandabi saas service!!!!

agent.train(docs=query.instructions)

It also stores every request (every agent chat question) to the pandabi service for some reason. This is really dangerous. The llm is clearly defined as:

llm = AzureOpenAI(
    api_token=os.getenv("API_TOKEN"),
    azure_endpoint=os.getenv("ENDPOINT"),
    api_version="2024-02-01",
    deployment_name="gpt-4",
)

[ArslanSaleem]

@flashtheman PandasAI API key is necessary if you want to train

[flashtheman]

@flashtheman PandasAI API key is necessary if you want to train

Yes, I see. But is that right? Is this part of the enterprise features? There is nothing in the docs about this. I thought only enterprise features and the Bamboo llm need the API key. It seems very confusing to me.

[gventuri]

@flashtheman that's a good point. We have updated the docs!

By default, it uses the BambooVectorstore (which relies on the free API token).
As an alternative, if you want to use local vector stores, they need an enterprise license for production.