Watchtower Free and Open Source Desktop App CookApps CookApps Marketplace

Overview

Watchtower is an open-source, AI-powered penetration testing automation CLI built on Python, LangChain, and LangGraph. Engineered for authorized security research and local DevSecOps auditing, Watchtower coordinates specialized Planner, Worker, and Analyst LLM agents to dynamically chain 23 real-world reconnaissance and vulnerability scanning tools and compile verified findings into clean PDF reports.

How It Works

Watchtower operates through structured, deterministic execution stages:

LangGraph Planner-Worker-Analyst Orchestration: Coordinates specialized Planner, Worker, and Analyst LLM agents in a cyclic graph to plan attack surfaces, run security scanners in parallel, and filter false positives.

Dynamic 23-Tool Security Arsenal: Auto-detects installed tools (nmap, masscan, httpx, subfinder, amass, nuclei, sqlmap, ffuf, dalfox, testssl.sh) on PATH with interactive or CI-ready headless selection.

Multi-Provider LLM & Authenticated Targets: Connects to OpenAI, Anthropic Claude, Google Gemini, OpenRouter, or custom HTTP endpoints via LiteLLM while passing custom session cookies and auth headers.

Local SQLite Memory & Automated PDF Reports: Logs every command execution, truncated vulnerability signal, and verified finding into pentest_memory.db to generate executive-ready PDF security reports.

Watchtower generated automated PDF penetration testing vulnerability report

Key Technical Features & Capabilities

Autonomous LangGraph Multi-Agent Loop: Separates engagement logic across a Planner (attack-surface strategy), Worker (parallel subprocess execution), and Analyst (stdout parsing, false-positive filtering, severity scoring).

23 Security Tools Auto-Discovered on PATH: Natively wraps nmap, masscan, httpx, whatweb, wafw00f, subfinder, amass, nuclei, sqlmap, ffuf, dalfox, testssl.sh, gitleaks, and more with interactive TUI toggling or --skip-ask-tools headless CI execution.

LLM-Agnostic Provider Architecture: Supports OpenAI, Anthropic Claude, Google Gemini, OpenRouter, Ollama local models, or any custom OpenAI-compatible HTTP endpoint via LiteLLM flags.

SQLite Persistent Engagement Memory & PDF Reports: Records every executed scan, smart-truncated observation, and finding in pentest_memory.db for instant on-demand PDF security report rendering.

Architecture & Technology Stack

Primary Languages: Python 3.11+

Frameworks & AI Runtime: LangGraph (Cyclic Agent State Graph), LangChain, LiteLLM, Pydantic Structured Outputs

Data Storage & Reporting: Local SQLite (pentest_memory.db), SQLAlchemy, ChromaDB, FPDF2 PDF Report Compiler

Supported Platforms: Linux (Native Recommended), macOS, Windows 10/11 via WSL2

System Requirements

Specification	Minimum Requirement	Recommended Specification
Operating System	macOS, Linux (Recommended), Windows 10/11 (WSL2)	Latest stable release build
Processor Architecture	4-Core x86_64 or Apple Silicon M1	8-Core Apple Silicon / Core i7+
Memory (RAM)	4 GB Unified Memory	16 GB+ High-Speed RAM
Disk Space	1 GB free SSD storage	5 GB+ free SSD storage

Installation & Setup Guide

2 Redit

Source Installation & Tooling Setup

bash

git clone https://github.com/fzn0x/watchtower.git
cd watchtower
python -m venv venv
source venv/bin/activate   # Windows: venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env       # Add OPENAI_API_KEY, GEMINI_API_KEY, or OPENROUTER_API_KEY
./install_tools.sh         # Install reconnaissance binaries (nmap, nuclei, httpx, etc.)

Installation & Setup Guide

Mở khóa ứng dụng để xem chi tiết hướng dẫn cài đặt, cấu hình môi trường và vận hành.

How to Use & Workflow

2 Redit

Configure LLM Provider: Populate your .env file with OpenAI, Gemini, or OpenRouter credentials (or pass --provider / --model CLI flags).

Launch Target Scan: Execute python -m watchtower.main -t https://your-staging-site.com (add --cookie or --header for authenticated endpoints).

Select Recon Arsenal: Toggle detected binaries interactively with Spacebar/Enter, or pass --skip-ask-tools for headless automation.

Generate PDF Report: Run python -m watchtower.main --report "pentest_report.pdf" to compile verified findings stored in pentest_memory.db.

How to Use & Workflow

Mở khóa ứng dụng để xem chi tiết hướng dẫn cài đặt, cấu hình môi trường và vận hành.

Troubleshooting & Common Issues

**Issue: Error does not support feature: structured-outputs with free/experimental OpenRouter models.**

*Cause*: Some lightweight models lack strict JSON schema output adherence expected by LangChain agents.

*Resolution*: Set OPENROUTER_MODEL_NAME in .env to anthropic/claude-3.5-sonnet, openai/gpt-4o, or google/gemini-1.5-pro.

**Issue: Edge WAFs return 403 Forbidden responses during noisy scanning.**

*Cause*: Aggressive tools such as masscan or ffuf trigger Cloudflare/CDN rate limits.

*Resolution*: Deselect brute-force modules from the startup checklist or scope tests strictly against staging/origin IPs.

Watchtower

Overview

How It Works

Key Technical Features & Capabilities

Architecture & Technology Stack

System Requirements

Installation & Setup Guide

Source Installation & Tooling Setup

Installation & Setup Guide

How to Use & Workflow

How to Use & Workflow

Troubleshooting & Common Issues

How to Use & Getting Started

How to Use & Getting Started

Ready to run Watchtower?

More in ai-automation

CodexClaw

Aside Codemode

AGBrowse