aduros / ai.vim Goto Github PK

in normal mode and insert mode, it is not entirely clear to me what region of code is being evaluated. Is it the line, the paragraph? Knowing this would be relevant to effectively use this plugin

Great plugin, this is really cool! Here are some suggestions how to make it even better:

Have a setup/config that allows to set things like:

• Check filetype and if it's go for example, add "in golang" to AI requests automatically.
• Which model(s) should be used for the different kind of requests
• Specify different keybinding than CTRL+A which is already mapped by default
• Choose between column-icon or virtual-text when waiting for response (maybe allow specifying the icon/text)
• Have option to read the API key from a file like ~/.openai or whatever as exporting isn't the most secure/best option especially for those who got public dotfiles.

Conversation-Mode where one can go back and forth with ChatGPT in a floating window and at the end have the last response parsed into the buffer.

Is it possible to support vim (non-neovim)?

how can i create a key mapping that asks for an input? For instance I have added this:

map("n", "<Leader>a", ":AI<CR>")
but the command should be :AI <type a question here>
I want to have the AI plugin input mode when I press <leader>A

<CTRL-A> is widely used in vim (e.g. incrementing numbers). I think this is not a good default.

Installed successfully; OPENAI_API_KEY is set.

When running :AI in vim, I get the following error:

ai.vim: You exceeded your current quota, please check your plan and billing details.

Everything works fine using the same account in a browser. I encountered the same error in a browser on occasion, but it was fixed by refreshing the window. Not sure how to accomplish the same here?

Error executing vim.schedule lua callback: ...unov/.local/share/nvim/plugged/ai.
vim/lua/_ai/openai.lua:83: Expected comma or object end but found T_END at chara
cter 336
stack traceback:
[C]: in function 'decode'
...unov/.local/share/nvim/plugged/ai.vim/lua/_ai/openai.lua:83: in funct
ion 'on_stdout_chunk'
...unov/.local/share/nvim/plugged/ai.vim/lua/_ai/openai.lua:14: in funct
ion <...unov/.local/share/nvim/plugged/ai.vim/lua/_ai/openai.lua:13>

I get this error after running <control + a>

Also is it possible to use gpt-4 if I pay for it with this?

Hey there love this plugin 🔥

I've spun off a hard-fork here that is based on ChatGPT and removes some functionality that you have here (diverges quite a bit from this main)

Changes introduced:

• Streaming completions with cancelling
• ChatGPT-only calls for completions & edits
• Removed some settings and loading state indicators that are no longer needed

Let me know if you'd like me to open a PR to merge those changes in 🙌

Can this be make to also work with regular vim?

Being able to define a custom location of a file containing the API key would be useful for those of us who made their dotfiles public.

The workaround right now is to set it via vim.env outside of the dotfiles repo, and symlink that file into the nvim config, which is unnecessarily many hoops to jump through just to get it set up.

Hello, could the plugin be update to use gpt-3.5-turbo?

thanks in advance, great tool you have here :)

The excommand interface is easy to use, but having a more powerful API could open up more advanced use cases.

When editing text by selecting and running with a prompt, it looks like the returned result has empty lines stripped out. This seems to be an issue in the plugin text replacement and not OpenAI.

If I am not interested anymore in the completion, or it has done its thing, I have no way to stop it. Would be great to be able to stop it, for example with ctrl+c. Would also be great not to have two concurrent completions running at the same time. Amazing plugin, thanks.

It's a large feature request but playing with https://cursor.so it seems very powerful.

AI refactoring, but instead of just making the editing having to undo / redo to figure out if it is what you want, the editor shows you an interactive diff where you can "accept all", "accept only this chunk" and their reject equivalents.

What do you think?

When using :AI from visual mode:

Error executing Lua callback: .../ygg/.local/share/nvim/lazy/ai.vim/lua/_ai/indicator.lua:41: Invalid 'end_col': out of range
stack traceback:
        [C]: in function 'nvim_buf_set_extmark'
        .../ygg/.local/share/nvim/lazy/ai.vim/lua/_ai/indicator.lua:41: in function 'create'
        ...e/ygg/.local/share/nvim/lazy/ai.vim/lua/_ai/commands.lua:52: in function 'ai'
        /home/ygg/.local/share/nvim/lazy/ai.vim/plugin/ai.lua:2: in function </home/ygg/.local/share/nvim/lazy/ai.vim/plugin/ai.lua:1>
Press ENTER or type command to continue

I added the plugin to Vundle list:

call vundle#begin()
Plugin 'aduros/ai.vim'
call vundle#end()

Then I ran :PluginInstall, and the repo does appear in ~/.vim/bundle
However using :AI returns:

E492: Not an editor command: AI

We currently have virtual text to show the response of ChatGPT in realtime and in the event of timeout, these received text are disposed of. It might be useful to retain, in the case where user don't have time to wait indefinitely for the entire response.

As title, sometimes ai.vim takes a really long time and returns error, sometimes it seems to break out of the loop and return a really long but repetitive result which I cannot use.

However, I might have already proceed on to other stuff and most likely do not need the result anymore, so I timeout with the user's preference might help a lot with the situation.