Strategies for HIPPA-compliant LLM integration?

[stephenlf]

I frequently use Marimo to transform data with sensitive health information. This data can't be sent to an LLM. As such, I have been too scared to experiment with Marimo's powerful LLM integrations.

I know there are a few solutions to this:

1. Use an LLM provider that will sign a BAA. (We haven't done this yet.)
2. Self-host a model. (Feasible, though we don't have the hardware for this. I'm sure I could convince my boss to buy something.)

I have a third approach in mind, though I don't know if Marimo supports this--

3. Configure Marimo's LLM integrations to not send over data frame contents.

Is it possible to do (3)? What would that look like?

---

Edit: OpenAI and Anthropic will now sign BAAs.

[mscolnick]

hi @stephenlf, currently it is not possible to configure the context sent to LLMs. but this is something we can consider for the future

[stephenlf]

We will look into HIPAA-compliant providers in the meantime.