talktoANYcode is my fork of TalkToCode, which adds support for any (most) programming languages, and thoroughly streamlines the preprocessing, including a cost estimate prompt. I rewrote the parsing for repos and files, as well as the path-handling code for file navigation in the frontend. TreeSitter is now used to split project source files into functions, and GPT-3.5 now handles completions.
Clone the git repository, get the submodules, and install the requirements:
git clone https://git.nategb.xyz/talk-to-any-code.git/ git submodule update --init --recursive pip install -r requirements.txt
create an object file for tree-sitter
get OpenAI token
python setup.py to create a CSV representing your repo that
has been processed by the OpenAI API
Usage: setup.py [OPTIONS] FILEPATH OUTPUT_CSV_FILEPATH
There is the option
--ignorefile to include an ignorefile
with directory names that you want to skip. One name per line.
You can also pass
--skip-summary to avoid generating
natural language summaries of your repo’s functions, because it can be
Now just start the Flask server with
python server.py CSV_PATH