I've used it a ton too and it feels quite polished. Lately I've seen codecompanion.nvim[1] around a lot and been wondering if someone has tried both, are there differences between these two in actual use that would make other one worth trying over the another?
I have tried to look into avante, but settled on codecompanion:
* codecompanion supports passing buffers, lsp and more. That seems to be really helpful when generating response.
* IMHO, codecompanion is more vimish (if that's the word)...
* Officially avante supports less models
* I have contributed to codocompanion GitHub models support (https://docs.github.com/en/github-models) so you can use some models basically for free with some limits and this is more than enough for daily usage from my experience (caveat that GitHub models are limited to 4000-8000 tokens so if that's problem for you then don't use it). I see avante does not support GitHub models.
I'm using Avante daily for several months and it is great but kot without it's quirks. Thanks for sharing - I will be testing this companion because it seems like a nicer DX with the @agent and #context mentions
I liked using Avante but unfortunately they seem to be favouring Claude primarily, and their Copilot implementation has issues to the point of printing warning messages. That's fine since it's their project; it just makes it less useful for my particular use case.
This is what I was hoping for when I saw this post! I hate to admit it, but I've recently switched to cursor as nvim experience without ai feels really slow now. Definitely going to try this!
Avante also has @quickfix (push quick fix list into the context) and my change to add @buffers (add open buffers to the context) was merged this morning.
My current dev setup is neovim + cursor. I'd love for it to one-day just be neovim (i.e. cursor's capabilities built into neovim). Seems that's getting closer. The 'Composer' feature is Cursor's 'Killer App' - the ability to instruct the LLM (which is context-aware) and have it generate the files/edits necessary, and be a bit 'agentic' (in that it self-critiques, occasionally backtracks, and does what a mid level dev would do in a day or two in 20 seconds).
It looks like it integrates directly with the Augment service [1] (and it's made by them). It sounds like Augment is some kind of a bespoke coding assistant that also does various RAG things to automatically get better context on large codebases.
Why I like Avante over others?
1. Active development. @Yetone, the creator, is very transparent and active.
2. Supports almost all the models. Add your API key for whatever you want to use.
3. The prompts are very well optimised. Plus the team keeps improving them.
4. The code `diffs` very well handled. It is easy to `apply` changes.
5. Support for `@file` feature (select multiple files) has made it 5x more powerful.
6. Very transparent. We can see what we are doing. Very less "magic".
7. It runs on demand: no auto-suggestion magic.