What's the difference between Posit Assistant (in Positron) and Positron Assistant?

[howardbaik]

I just gave the Posit Assistant a try in the May release of Positron and was thrilled to see it included. While I knew it had been available in RStudio, I never got around to trying it there since I've fully transitioned my workflow over to Positron.

I'm having a hard time understanding what distinguishes Posit Assistant (in Positron) from Positron Assistant. From my 5 minutes of playing around, they are both assistants that can run code in the console, read/edit code in the editor, etc. They both allow you to use the new Posit AI as a model provider. In my opinion, Positron Assistant has a better UI than Posit Assistant, so I'm having a hard time switching to Posit Assistant (in Positron). Maybe the difference is that Posit Assistant (in Positron) is just a hybrid of Positron Assistant and Databot.

Given that I use Anthropic Claude as my model provider, what would be a reason for me to switch to Posit Assistant (in Positron)?

[jmcphers]

You're right to note the similarities. In upcoming releases we are going to continue to invest in making Posit Assistant better and more capable. Positron Assistant -- which is based on Microsoft's Copilot UI and APIs -- will wind down and revert to its baseline Copilot experience. Copilot does support BYOK but won't have all the data science integrations Posit Assistant comes with.

If there are things about the Positron Assistant UI that you'd like to see in Posit Assistant (and/or Positron) we'd love to hear about them!

[jcheng5]

Hi Howard, I'd be interested to hear what aspects of the Positron Assistant UI you prefer to the Posit Assistant UI! It's our goal to make Posit Assistant feel like an upgrade for almost everyone.

[jeffkeller-einc]

I prefer the way Positron Assistant presents code edits, allowing me to keep, undo, or modify the edits myself before accepting them. With Posit Assistant, it seems to be all or nothing and is presented within the chat pane itself, which isn't very wide most of the time I'm using it, resulting it a lot of horizontal scrolling.

[howardbaik]

Hi @jmcphers and @jcheng5, I've only been using Posit Assistant for a couple of days, but here are a few initial observations. I'll add to this list as I come across more.

1. The user message box has excessive whitespace:

By comparison, the user message box in Positron Assistant has more appropriate spacing:

I assume this box is styled with the standard CSS box model, in which case reducing the padding (or possibly the border) around the content should resolve it.

2. The icons in the top action bar are too small and low-contrast:

Positron Assistant's icons are larger and more clearly defined:

3. The gold font color in the "Can Posit Assistant use X" question boxes seems a bit heavy. Matching the font color to the red/orange of the border might be a cleaner choice:

[jcheng5]

Thank you!!

[moa-incom]

I will be a +1 for the small text/buttons and contrast/border differences; that also includes the actual "Ask Posit Assistant..." textbox, that is very dim in Posit Assistant.
Also, I find it hard(er) to include context. Let's say I want to include context from a file I don't currently have open, in Posit Assistant I need to open the file, whereas in Positron Assistant I could just find the file via the "+" (Add Context...).
Another context difference that I find slightly annoying is the fact that all files currently open are per default included, and to unselect I need to open a pop-up bar.
Lastly, I'm having some trouble setting the default models and web search allowance etc. in Posit Assistant.

EDIT: I just came across one more thing, I like how in Positron Assistant you can decide whether it's in ask/agent/edit mode, whereas in Posit Assistant it seems to be only in Agent mode.

[howardbaik]

To piggyback on your first point, I would say that all the icons / text / buttons in Posit Assistant are awkwardly small.

[jcheng5]

Thank you both for the feedback. In Posit Assistant, you don't tend to manage the context manually by including files, instead you lean on the model's ability to find what it needs on its own. And if you do want to make sure it looks at a specific file, you don't need to open a file to make it included; instead, mention the filename (you can type an @ character to have the UI help you find the file) and the agent will go read (parts of) it.

This may sound weird if you're very used to Positron Assistant (or Copilot) but it's the way almost all modern coding agents work. If you can't get used to it, I'm happy to hear more about what you're doing.

We're working on the button sizes, colors, etc. Thanks!