Positron assistant feedback

[gshotwell]

Great work on implementing an assistant. I've been a long term Cursor user and really wanted something like this for Positron. I wanted to share a few thoughts about this which might be helpful for this tool:

1. Add a "use cursor shortcuts" checkbox to the assistant setup. Most people will have a very strong CMD+L muscle memory at this point and it's good to make it easy for them to keep that shortcut. I actually had quite a hard time opening the chat window.
2. Maintain data science documentation embeddings/MCP servers. The best feature of Cursor IMO is that you can attach documentation to the chat query more easily. This feature has really degraded over the past while with Cursor, and was never very good for open source tools. Being able to call @Shiny to attach the shiny docs would be a big differentiator.
3. Marketing: One thing I've done a lot with Cursor is asked an LLM to validate my understanding of diagnostic plots in new statistical areas. For example I was working on a multi level model and didn't quite understand some of the default plots, attaching the picture to a chat window is quite helpful
4. Marketing: it's worth following the cursor subreddit because people over there are increasingly pissed off at Cursor for many of their features. They've been prioritizing agent workflows which has led to degraded performance and lots and lots of expensive tool calls. I think you can position Positron as a transparent chat client for data science and make a lot of headway.

[jthomasmock]

Thanks Gordon!

1. Good callout! Do you also find yourself missing Cursor rules or other custom prompt enhancements/project context? Are there other text-based prompt enhancements you'd like?
2. We do have some interest in coordinating w/ MCP servers, but it's also good to hear about interest with proper embeddings/RAG for popular data science libraries/documentation. There's also some work on help-page documentation integration.
3. Totally agree -- the unique UI/tools are something we think will really help with interactive analysis and development.
4. Interesting, thanks for the context!

[gshotwell]

I do use cursorrules files from time to time, and that's probably something that's good to integrate. Cursor's economic model is to markup the calls to the LLM which means that over the last 6 months or so a lot of their workflows have moved towards pushing people to use many many calls which IMO has made the product a lot worse. For example you used to be able to tell the LLM to look at your codebase with @codebase but now every question will result in multiple codebase searches even if you didn't really ask it to. A lot of the changes on Cursor's side have been quite opaque and frustrating.

Since Positron has a different economic model, I think you all could focus on making the LLM part of the product transparent, stable, and customizable.