IntelliJ IDEA Plugin: LLM-powered Code Analysis

This IntelliJ IDEA plugin provides seamless integration with powerful code analyzers:

• Large Language Models (LLM)
• Semgrep
• CodeQL (only with --build-mode=none)

The plugin allows you to analyze code files directly from the Project View using these tools. It simplifies the setup process by leveraging Docker for running CLI-based analyzers and supports customizable configurations.

Features

• Code Analysis Options: Analyze code with LLM, Semgrep, or CodeQL by right-clicking on any file in the Project View and selecting:
  • Analyze Code -> LLM
  • Analyze Code -> Semgrep
  • Analyze Code -> CodeQL
• Docker Integration: Use Docker to run CLI versions of code analysis tools.
• Flexible Configurations: Easily configure API keys, model settings, and Docker hosts via a JSON configuration file.

Installation

1. Clone the repository:

   git clone <repository-url>
   cd llm-analyzer-plugin
   

2. Install required docker images:
   • For CodeQL (3.79gb):

   docker build -t codeql docker/codeql
   

   • For Semgrep (580mb):

   docker pull semgrep/semgrep:latest
   

   Semgrep image may also be downloaded automatically when you use it in the first time.
3. Build and install the plugin:
   • Open the project in IntelliJ IDEA.
   • Set configuration variables (see below)

Usage

1. Click "Run Plugin" in your IDE.
2. In opened testing IDEA environment open any java project.
3. Right-click on a java file in the Project View.
4. Select Analyze Code and choose the desired analyzer.
5. View the analysis results in the corresponding files.

Configuration

The plugin requires a configuration file located at resources/settings.json with the following structure:

{
  "apiKey": "<your-api-key>",
  "baseUrl": "<api-base-url>",
  "model": "<llm-model-name>",
  "dockerHost": "<docker-server-host>"
}

On Windows, in Docker Desktop you have to go to Settings -> General and select "Expose daemon on tcp://localhost:2375 without TLS" option.

Parameters

• apiKey: Your API key for accessing LLM services.
• baseUrl: The base URL for the LLM API (default is Gemini API URL).
• model: The LLM model to use (default is Gemini 2.0 Flash).
• dockerHost: The Docker server address for running CLI analyzers. tcp://localhost:2375 (default) on Windows, unix:///var/run/docker.sock on Unix-like.

Dependencies

• Docker Desktop or any other Docker host: Required for running CLI tools.
• Docker-Java: Used for Docker communication.
• OpenAI4J: Provides integration with LLM APIs.