CRANE – Code Review AI Network Engine (Master Thesis Project)

CRANE is a research-driven system designed to explore the potential of Large Language Models (LLMs) in automating and enhancing the code review process through multi-agent collaboration. At its core, CRANE transforms the traditional code review workflow into a network of LLM agents that engage in structured conversations, each playing a specialized role in reviewing, critiquing, and ultimately resolving code review comments (CRs).

This project investigates a novel approach to prompt engineering where LLM agents not only analyze code and comments but also communicate with each other, mimicking a team of human reviewers. Their dialogue is then synthesized into actionable feedback that another LLM agent or a human can use to revise or improve the original code.

---

Getting Started

1. Clone the Repository

git clone https://github.com/LucoMoro/CRANE
cd CRANE

2. Install Dependencies

Make sure you have Python installed (preferably 3.12.9), then:

On Ubuntu/Linux and macOS

Important: Before installing, open requirements.txt and remove the following line:

pywin32==310

This package is Windows-only and will cause installation errors on Linux-based systems. Then:

python -m pip install -r requirements.txt

On Windows

python -r requirements.txt

3. Configure Environment Variables

Create a .env file in the root of the project and add the following variables:

HUGGINGFACE_API_KEY=your_huggingface_api_key
OPENAI_API_KEY=your_openai_api_key
PINECONE_KEY=your_pinecone_key
BASE_PATH=./conversations
DATASET_PATH=./dataset

4. Running CRANE

A sample Change Request (CR) has already been included in the dataset folder.
This allows you to quickly test whether the model and environment are working correctly before adding your own CRs.

On Ubuntu/Linux and macOS

Ensure model paths in the code do not include "../" — update them to be relative (e.g. "models/..." instead of "../models/..."). To run the project:

python3 -m network.main

On Windows

You can keep "../" in model paths if they work for you. To run the project

python main.py

---

Project Structure

crane/
├── agents/              # Code for individual LLM agents
├── network/             # Multi-agent coordination logic
├── conversations/       # Saved agent conversations and interactions
├── dataset/             # Folder containing the CRs
│   ├── snippets/        # CRs' code snippets
│   └── tasks_description/ # CR's Task descriptions
├── utils/               # Helper functions
├── prompts/             # Prompts used by the LLMs
├── .env                 # Environment variables (not committed)
├── requirements.txt     # Python dependencies