DEPRECATED: Try Mentat instead! https://github.com/AbanteAI/mentat
Give your python scripts regenerative healing abilities!
Run your scripts with Wolverine and when they crash, GPT-4 edits them and explains what went wrong. Even if you have many bugs it will repeatedly rerun until it's fixed.
For a quick demonstration see my demo video on twitter.
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
cp .env.sample .env
Add your openAI api key to .env
warning! By default wolverine uses GPT-4 and may make many repeated calls to the api.
To run with gpt-4 (the default, tested option):
python -m wolverine examples/buggy_script.py "subtract" 20 3
You can also run with other models, but be warned they may not adhere to the edit format as well:
python -m wolverine --model=gpt-3.5-turbo examples/buggy_script.py "subtract" 20 3
If you want to use GPT-3.5 by default instead of GPT-4 uncomment the default model line in .env:
DEFAULT_MODEL=gpt-3.5-turbo
You can also use flag --confirm=True which will ask you yes or no before making changes to the file. If flag is not used then it will apply the changes to the file
python -m wolverine examples/buggy_script.py "subtract" 20 3 --confirm=True
| env name | description | default value |
|---|---|---|
| OPENAI_API_KEY | OpenAI API key | None |
| DEFAULT_MODEL | GPT model to use | "gpt-4" |
| VALIDATE_JSON_RETRY | Number of retries when requesting OpenAI API (-1 means unlimites) | -1 |
This is just a quick prototype I threw together in a few hours. There are many possible extensions and contributions are welcome:
- add flags to customize usage, such as asking for user confirmation before running changed code
- further iterations on the edit format that GPT responds in. Currently it struggles a bit with indentation, but I'm sure that can be improved
- a suite of example buggy files that we can test prompts on to ensure reliability and measure improvement
- multiple files / codebases: send GPT everything that appears in the stacktrace
- graceful handling of large files - should we just send GPT relevant classes / functions?
- extension to languages other than python
