patchllm 0.1.0

Lightweight tool to manage contexts and update code with LLMs

About

PatchLLM lets you flexibly build LLM context from your codebase using search patterns, and automatically edit files from the LLM response in a couple lines of code.

Usage

Here's a basic example of how to use the Assistant class:

from main import Assistant

assistant = Assistant()

context = assistant.collect(config_name="default")
>> The following files were extracted:
>> my_project
>> ├── README.md
>> ├── configs.py
>> ├── context.py
>> ├── main.py
>> ├── parser.py
>> ├── requirements.txt
>> ├── systems.py
>> └── utils.py

assistant.update("Fix any bug in these files", context=context)
>> Wrote 5438 bytes to '/my_project/context.py'
>> Wrote 1999 bytes to '/my_project/utils.py'
>> Wrote 2345 bytes to '/my_project/main.py'

You can decide which files to include / exclude from the prompt by adding a config in configs.py, specifying:

• path: The root path from which to perform the file search
• include_patterns: A list of glob patterns for files to include. e.g [./**/*]
• exclude_patterns: A list of glob patterns for files to exlucde. e.g [./*.md]
• search_word: A list of keywords included in the target files. e.g ["config"]
• exclude_extensions: A list of file extensions to exclude. e.g [.jpg]

Setup

PatchLLM uses LiteLLM under the hood. Please refer to their documentation for environment variable naming and available models.

License

This project is licensed under the MIT License. See the LICENSE file for details.