Post

If you have sensitive data, how (if at all) do you use AI? Just chat with LLMs on the web (so they don't have access to your data)? Anonymize your data before using it with Claude Code (or similar)? Not use AI at all? #rstats

Dirk Eddelbuettel

@eddelbuettel.com

Local models. Nothing leaves your machine. LM Studio or Ollama run them, also as a service that e.g. RStudio or positron can hook in. Open weights such as qwen3.8 are pretty good.

David Keyes

@dgkeyes.com

When you say that RStudio or Positron can hook into local models, how does that work exactly?

Dirk Eddelbuettel

@eddelbuettel.com

Both LM Studio and Ollama run / can run as a service that gets accessed via a localhost url and port (it's actually an OpenAI compatible setup). The IDEs (just like any other service: pi, openclaw, ...) connect to that under their settings. For positron you may have to edit a JSON file.

28/9/26, 5:04 pm0 reposts0 likes

David Keyes

@dgkeyes.com

Ah gotcha. Didn’t realize this was through Posit Assistant.

Dirk Eddelbuettel

@eddelbuettel.com

The key is that this is a well-understood interface offered by the 'engines' (such as Ollama or LM Studio or ....). Posit Assistant is but one client, it happens to be inside the IDE. I happen to use OpenCode (in the terminal) and / or pi much more for it.