Enterprise Mobile Pentesting

Automating Android Audits via
Cognitive AI Orchestration

APPredator integrates local decompilers with cognitive language models to eliminate false positives in Android SAST and generate sandboxed-tested proof-of-concept exploits.

Android Security Workspace

Welcome to APPredator

Static analysis, LLM-assisted review, and hardening workflows in one console. Upload an APK, tune your RAG prompts, and track findings from scan to report.

Toolchain

Java Apktool JADX

LLM Engine

deepseek

Scan Jobs

0 active
Quick Access
๐Ÿ”

New scan

Upload an APK or XAPK and queue static + LLM analysis.

๐Ÿ“ˆ

Live dashboard

Stream logs, severity charts, and partial findings.

๐Ÿ“

RAG & prompts

Edit retrieval context and rule templates.

Designed for High-Precision Vulnerability Auditing

APPredator solves the false positive problem in Android static testing by replacing raw signatures with localized context evaluation.

Precision Triaging

Rather than flagging simple APIs, APPredator feeds decompiled Java bytecode and class properties into advanced LLMs to semantically analyze whether the code path is exploitable.

Local-First & Private

Static pre-filtering, decompilation, and call-graph generation occur entirely on your local machine. Code snippets are isolated before being dispatched to your LLM API of choice.

Exploit Prototyping

A confirmed vulnerability is paired with a ready-to-run exploit script. The platform auto-validates these scripts locally inside a secure runtime sandbox before outputting.

Inside the Scanning Engine

The step-by-step pipeline of how APPredator converts raw binaries into actionable vulnerability reports.

1

Local Decompilation

The JADX wrapper converts Dalvik Executable (dex) binaries to clean Java source files. Concurrently, Apktool extracts binary resources, layout assets, and smali registers.

2

Heuristic Pre-Filtering

The local CodeFilter scans class trees to identify high-value targets (e.g. WebViews, SharedPreferences, SQLite databases) and compiled regexes, optimizing API overhead.

3

Call-Graph Dependency Resolution

Traces method references (Xrefs) to construct a localized dependency map. This ensures the cognitive AI is injected with context-dense code execution blocks.

4

Cognitive LLM Auditing

Queries cognitive LLM providers utilizing prompt templates enriched with dynamic OWASP MASVS compliance mappings and Android manifest component context.

Integrated Technologies & Runtimes

Built-in support for industry-standard reverse engineering runtimes and cognitive language models.

Decompilers

JADX & Apktool
JVM Native

AI Engine

Google Gemini
2.0 Flash / Pro

AI Engine

DeepSeek
DeepSeek-V3 / R1

Local AI

Ollama
Llama3 / Codegemma

Quick Start Guide

Follow these standard terminal inputs to spin up the local console on your host system.

bash - apppredator@localhost
# Prerequisites: Java JRE 21+ and Python 3.11+
# 1. Clone repository and install in editable mode
git clone https://github.com/joelindra/APPredator.git
cd APPredator
pip install -e .

# 2. Start the interactive console LLM configuration wizard
apppredator settings setup

# 3. Boot the local Uvicorn FastAPI server and open the Web UI
apppredator web
Authorized testing only โ€” use on apps you own or are permitted to assess.
Created by Joel Indra ยท GitHub ยท hadesxploit.com