This looks amazing!  Glad to see MV tools not just in the 21st century but catching up with the bleeding edge :-)

I work mostly with banks and they are very sensitive to private information leaving the network.  Is there more detail on the models used and what is shared with the Rocket Software on AWS?

• Program delivery: Hosted by Rocket Software in AWS and free with an MV data server (UniVerse, UniData, D3, jBASE, or OpenQM) serial number on active maintenance.

Hello Ian -

Sorry for the late reply. 

First, the MV Developer Assistant is the result of fine tuning  a base LLM, Meta's Llama 3.1, with MV specific training data. 

Second, the assistant is hosted in AWS, and is a basic web server reverse proxy. The LLM is running on the same server. When you submit a question from VSCode (using the Continue extension), your request is sent to the AWS machine, processed by the LLM, and the response is returned. Nothing is stored on the server, except typical web server logs with generic information about your request (response status, request length, response length, response processing time, origin IP address and your API key). The request and response content is not logged.

When you use the autocomplete feature (fill in the middle), code from the program you are editing is sent to the AWS machine. This would include several lines before and after the cursor. If you followed the instructions provided for configuring the autocomplete model in the Continue extension (and provided that any updates to the Continue extension do not affect the meaning of this configuration), no other data is sent to the server. As before, only generic request information is logged. 

See the documentation for information about the autocomplete feature. The Continue extension is not able to restrict autocomplete to only BASIC code, so it will attempt to autocomplete any document you are editing in VSCode. We recommend disabling Continue autocomplete while you are editing other types of files.

Hope this helps.

Thanks, Pete