I recently learnt about https://github.com/Robitx/gp.nvim and found it's really close to what I was trying to build with a bunch of great features already implemented. I was using this plugin but not adding the features I wanted, archiving so it's clear I won't add any new features.
This is a plugin for interacting with the openai chat api. It's distinctive feature is that the conversation happens in a regular editable buffer so that it plays nicely with other vim plugins and so the user can revise the conversation history before continuing it.
Requires plenary.nvim, curl, and a openai api token in the environment variable OPENAI_API_TOKEN.
This plugin is very work in progress, as in it only just works. There's no docs yet, read the source code (currently <300 lines in one file). Contributions and ideas welcome. Unless you really like exactly what I'm making here I'd recommend one of these two plugins instead:
- Handle error states better, currently if not online it just says loading forever, not sure what happens if openai's api is down or erroring
- Using virtual text for error and loading state (and maybe expose a function so it could be put in a status line)
- Use the streaming api so tokens can be added to the buffer immediately (1 token per ~100ms instead of the whole lot in 3-10 sec)
- Is it possible to cancel a streaming response part way through, so we can shut it up when it's waffling? That would be sweet
- Add some formatting and/or concealing to the section markers
- Opt-in debug logging that saves all
curlrequests and responses to file
tests/init.lua will download plenary to .tests/ on first run.
Assuming you have just installed, run tests with: (if you don't have just installed read the justfile for the command to run)
# run all tests
just test
# run one test
just test tests/files_spec.lua
# run tests on any lua file change (requires entr to to be installed)
ls **/*.lua | entr just test