Features matrix with Cursor

[tlvenn]

I think it would be great it we could somehow try to maintain a feature matrix with an AI vscode editor like Cursor, i would be tempted to add Cline & Windsurf but that might add unnecessary burden.

The idea would be for people who are not so familiar with the latest dev on AI, AI Agent, Agentic Workflow, MCP, Rules, etc to grasp how CodeCompanion is faring against those editors.

There is an increasing pressure that if you don't use one of those tools, you are essentially missing out and I am hoping such matrix would help people understand better where CodeCompanion stands.

[olimorris]

...this sounds like an ideal community contributed docs table 😉. I hear you and totally the get the sentiment. I'm full time (in my spare time) on CodeCompanion so the first I ever hear about other editor features are through the community.

As with anything that relates to product marketing, I think this could be like comparing apples and oranges. The features section of the README is designed to give the reader a high level summary of what the plugin can do. I think a more impactful section would be on how to use the plugin to achieve common outcomes. This was something that I've earmarked the usage/introduction page to be, but haven't had the time to flesh it out in more detail.

[tlvenn]

I need to get more familiar with CodeCompanion first but then would be happy to try to provide this. Thanks a lot for this amazing plugin @olimorris !

[olimorris]

Feel free to post hear, tag me if things don't make sense. I get to learn about the plugin based on how you learn so feel free to share your thoughts (good and bad). Anything I can do to streamline the docs, make it easier for new users to pick up etc.

[ilan-schemoul]

The number of features in this plugin or equivalent plugin/IDEs is quite high and it's non trivial to have an up to date clear vision of how different this is from alternatives.
Reading the docs both seems very similar but it's hard to really understand the differences.
I'll take an example: one of the bullet point of Cursor website is "Knows your codebase" it claims it can use your codebase to give answers. Codecompanion talk about workspaces it seems similar. But in terms of concrete details: do both use my entire codebase ? Do I have to set things up ? Is one more feature advanced ?
It's just one example but it shows it's hard for an outsider to compare.

Especially that things move at a breaknace speed around AIs currently. Two years ago I tried some nvim plugins and it sucked, completions were dumb and obnoxious and the plugin to chat with AIs was very buggy and AIs were underwhelming. As I understand now AIs became good at coding under strict supervision (Claude 3.7 for example) and "agents" allows to have a workflow where AIs can make plans, ask if you accept/refuse code, ask feedback, run the test itself etc. So as I understand things my value as a dev will decline quickly if I don't get up to date with new tools.

So it would be awesome if anyone would provide a table comparing their in depth experiences with cursor vs codecompanion

[olimorris]

I think that's a great point and a great idea.

I would be more than happy to feed into this. Perhaps unsurprisingly, I have very limited knowledge of Cursor and other Neovim plugins so won't be very good at filling in the answers for those.

I think Cursor and Aider would be great comparators.

[tlvenn]

I am exploring Cursor & Windsurf, there are a lot of things going on from the system prompt, to the exposed tools and how they try to feed the llm as much context as they can as best as they can.

Looking at extracted system prompts give a little bit of an idea of what is actually going on, for example for cursor

https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools/blob/main/Cursor%20Prompts/cursor%20agent.txt

Regarding the tooling, i have the feeling that with MCP becoming the standard, we will be able to plug pretty much any kind of tools in any system so this will be less of a differentiator.

[ilan-schemoul]

what is next edit suggestion ? I couldn't find this on Cursor doc.
Is it different than using cmp to apply suggestions ?
EDIT: oh okay they call it "tab", it's similar to Github copilot but has more context. SuperMaven indeed seems similar, I'm wondering if it's as good and has as much context tho

[e2r2fx]

Its main appeal is doing quick local refactors.

TLDR; Its either expensive and precise, or cheap and inaccurate. Tough to strike a balance there without investing in optimizing a model.

For achieving this it would take a cheap optimized LLM continuously submitting file diffs and coming back with a data structure of suggested buffer edits, then displaying those edits with mini.diff, and being able to accept them with a key bind. I've played around with it, but the thick of it is cost and precision.

[ilan-schemoul]

What did you use ? SuperMaven, cursor, copilot ?
If you have used multiple tools I'd be very interested in any feedback comparing those.
I think at some point I'll just try multiple tools to form my own opinion because surprisingly I can't find a lot of tdata to compare codecompanion vs cursor etc.
Actually I just learned supermaven got bought out by Cursor and it seems like they stopped updating their tool :(

[e2r2fx]

Well so far I've used the ones you mentioned plus in no particular order:

• Codeium for autocomplete (eventually switching to SuperMaven since its way faster and useful)
• Avante (another plugin trying to emulate the cursor experience in nvim, I very much prefer CodeCompanion's approach)

[e2r2fx]

I think that for trying AI coding specifically cursor has the best DX if you are good at configuring VSCode (I mean this because I have an intricate setup in neovim that would take me a while to reproduce in there :) and just want to use the tool as is. CodeCompanion simply offers the best experience for Neovim, for the features that matter to me:

• Context Management
• Agents
• Extensibility
• Inline operations

[tlvenn]

Yes official rules support would be super nice, there is a contribution in the discussion, did not have a chance to play with it yet. For the "tab" feature, I use SuperMaven which works quite well.

[tlvenn]

For leaked prompts, there is also this repo: https://github.com/jujumilk3/leaked-system-prompts