Jadaix Plugin for JADX

A JADX plugin for interactive code analysis using Large Language Models (LLMs). Provides dynamic code analysis, security assessment, malware analysis, and contextual understanding of decompiled Android applications.

Features

Analysis Types

• Single Class Analysis: Analyze individual classes
• Method Analysis: Detailed analysis of specific methods with context

Hierarchy Analysis

Performs deep analysis of class relationships by:

1. Direct Relationships

   • Analyzes superclass inheritance chain
   • Identifies implemented interfaces
   • Maps inner classes and their relationships

2. Usage Analysis

   • Scans for direct references in code
   • Identifies classes that use the target class
   • Traces method invocations and field usage
   • Maps package-level relationships

3. Package Scanning

   • Uses package prefix matching for efficient scanning
   • Optimizes relationship finding for large codebases
   • Applies whitelist/blacklist patterns for focused analysis

4. Memory Efficient Processing

   • Analyzes classes on-demand without storing relationship maps
   • Uses pattern caching to optimize repeated checks
   • Maintains minimal memory footprint for large apps (400,000+ classes)
   • Limits analysis depth to 2 levels to prevent recursion issues

5. Example Hierarchy Flow:

TargetClass
│
├── SuperClasses
│   ├── DirectSuperClass
│   └── IndirectSuperClass
│
├── Interfaces
│   ├── DirectInterfaces
│   └── IndirectInterfaces
│
├── Usage
│   ├── ClassesUsingTarget
│   └── ReferencingClasses
│
└── Related
    ├── SamePackageClasses
    └── DirectlyReferencedClasses

6. Performance Tips
   • Use specific package patterns in whitelist
   • Blacklist common framework packages
   • Start with key classes in the application

Analysis Modes

• Security Analysis: Identifies potential security vulnerabilities, focusing on concrete findings in the code
• Malware Analysis: Analyzes code for potential malicious behaviors and patterns
• Custom Analysis: User-defined analysis with custom prompts

LLM Support

• OpenAI (GPT-4)
• Claude (Anthropic)
• Mistral
• Self-hosted Mistral

Installation

1. Build the plugin:

./gradlew clean build

2. Install the plugin in JADX:

• GUI: Preferences → Plugins → Install plugin → Select built jar
• CLI: jadx plugins --install-jar jadaix-plugin.jar

Configuration

LLM Settings

• Model Selection (GPT-4, Claude, Mistral)
• API Keys
• Mistral Self-hosted Endpoint
• Analysis Type Selection (Security/Malware/Custom)
• Custom Prompt Configuration

Analysis Filters

Default blacklist patterns (can be customized):

android.*, androidx.*, com.google.*, java.*, javax.*, 
kotlin.*, dalvik.*, org.json.*, org.xml.*, 
org.w3c.*, org.apache.*, sun.*

Chat Interface Features

• Syntax highlighting for code blocks
• Markdown rendering for explanations
• Copy button for responses
• Persistent chat history
• Tokyo Night theme (yes its riced)

Performance Notes

• Hierarchy analysis performance depends on codebase size
• Use blacklist patterns to exclude irrelevant packages
• Memory efficient, suitable for large applications
• Manages token limits automatically
• Be careful with token usage, Hierarchy analysis will burn through them quickly set context limits!
• Chat context is maintained like a standard chatbot (each message will include the full chat context plus the new message, so account for this with token usage)