Skip to content

Home

Jump to bottom
Lukas Rump edited this page Nov 21, 2024 · 5 revisions

Effortlessly Get Code Reviews from Large Language Models!

CRLLM is a powerful command-line tool that enables developers to get code reviews from state-of-the-art Large Language Models (LLMs). Whether you want to use Ollama for locally running LLMs or connect to cloud services like ChatGPT, Hugging Face, and Azure, CRLLM has you covered. Improve your code quality, catch potential bugs, and receive AI-driven suggestions to enhance your development workflow. Get recommendations on best practices, bug-prone areas, and potential refactoring. Learn from the suggestions how to improve your code, gaining new insights and techniques. Reduce the need for back-and-forth during human code reviews by catching more issues upfront.

🚀 Features

  • Flexible LLM Options: Use Ollama to run models locally or leverage APIs from cloud providers like OpenAI, Hugging Face, and Azure.
  • Comprehensive Code Review: Get quality feedback on code, including recommendations for readability, efficiency, and best practices.
  • Developer Productivity: Integrate seamlessly into your existing development environment to speed up code review processes.
  • Privacy Control: Choose between local or cloud-based solutions depending on your privacy needs and computational power.
  • Multi-Language Support: Review code written in various programming languages (e.g., Python, JavaScript, Java, etc.).
  • Git Support: Get Reviews for your Git changes or differences between branches.

🛠️ Installation

To get started with crllm, follow these simple installation steps:

Prerequisites

Install from PyPI

pipx install crllm

Install from GitHub

pipx install git+https://github.com/lukasrump/crllm.git

🌐 Configuration

CRLLM supports multiple backends for LLM code reviews. You can configure it by adding an configuration file crllm_config.toml in the root of your project. To initialize your project you can use

crllm -i .

This command guides you through the most important settings. This TOML configuration file is splitted in four main sections:

[project]

  • description: Short project summary.

[crllm]

  • loader: Mechanism to load the source code, "git" by default.
  • provider: LLM provider, "ollama"by default.
  • git_main_branch: Specifies the main git branch, default is "main".
  • git_changed_lines: If true, only reviews changed lines.

Loaders

  • file: Code review for a single source code file
  • git: Reviews all changed files in the git repository
  • git_compare: Reviews the difference between the current git branch and the git_main_branch

[model]

The model settings depend on the provider. The model settings are the same as those of the LangChain ChatModels. Per default crllm tries to use a locally installed ollama instance with qwen2.5-coder:32b.

Ollama Local Setup

  • model: Specifies the model to use, e.g "llama3.1". Make sure that you pulled that model before you use it.

OpenAI API

  • model: Specifies the model to use, e.g "gpt-4o".

In addition you have to define the api key in your environment (.env)

OPENAI_API_KEY=your_openai_api_key

Hugging Face API

  • repo_id: Specifies the repository to use, e.g "HuggingFaceH4/zephyr-7b-beta".
  • task: Specifies the task, e.g "text-generation".
HUGGINGFACEHUB_API_TOKEN=your_huggingface_api_key

Azure OpenAI

  • azure_deployment: Specifies the deployment to use, e.g "gpt-35-turbo".
  • api_version: Specifies the api version to use, e.g "2023-06-01-preview".

In addition you have to define some variables in your environment (.env)

AZURE_OPENAI_API_KEY=your_azure_api_key
AZURE_OPENAI_ENDPOINT=https://your-endpoint.openai.azure.com

[prompt]

  • template: Override the prompt template that is used (optional).

✨Usage

CRLLM is designed to be easy to use right from your terminal. Below are some examples of how you can leverage the tool.

To perform a code review for a file or GIT repository run:

crllm path/to/your/repo/or/file

Enabling RAG Support

To enhance code reviews with source context, enable RAG (Retrieval-Augmented Generation) in crllm_config.toml:

[rag]
enabled = true
embedding_model = "all-minilm"      # Specify the embedding model
src_path = "./"                     # Define the root path of your source code
src_glob = "**/*.py"                # Use glob patterns to match source files (e.g., Python files)

Ignore files

CRLLM supports a .crllm_ignore file to exclude specific files and directories from code reviews. This is similar to .gitignore but specific to CRLLM's code review process.

Clone this wiki locally