Skip to content

Getting Started

Fresh

Source: README.md, bible/understand-everything.md

What Ghost Engine is

Ghost Engine is a second-generation antidetect browser automation engine. It runs on port 4700 on a Windows 11 machine. Its job: make automated browser sessions look identical to a real human sitting at the keyboard.

The first generation, Mini Engine V1 on port 4600, scores 98 of 100 on fingerprint detection tests. V1 handles daily SEO operations: launching persona browsers, running CTR campaigns, warming accounts, posting to social platforms.

Ghost Engine exists because V1 has a ceiling. Platforms like Reddit, LinkedIn, and Facebook check whether input events came from real hardware. V1 uses Chrome DevTools Protocol (CDP) to send clicks and keystrokes, and CDP produces isTrusted: false on every event. That is a dead giveaway. Reddit shadow-bans accounts that pass every fingerprint test because it can see the clicks are not real.

Three new capabilities

Ghost Engine fixes this with three capabilities V1 cannot have:

  1. OS-level input. On desktop, mouse and keyboard events are sent through Windows user32.dll SendInput. The OS injects them into the input stream. Chrome sees them as real hardware events: isTrusted: true.

  2. ADB input. On Android (physical devices or Waydroid emulators), touch and keyboard events are sent through Android Debug Bridge. Android treats ADB input as real touch: also isTrusted: true.

  3. Vision navigation. Instead of finding buttons by CSS selector (which requires CDP calls that leave traces), Ghost Engine screenshots the page, sends the image to a vision-language model (Claude, Ollama, or OpenAI), and gets back pixel coordinates. Zero DOM interaction.

The cold-start

Ghost Engine is mini engine V2: a unified desktop plus mobile antidetect browser cloned from V1 and extended with three I/O backends. It runs on port 4700 alongside V1 on 4600, separate data dir, separate API key, no interference.

The four new modules are:

  • os-input.mjs: Windows user32.dll P/Invoke via PowerShell plus inline C#
  • adb-input.mjs: humanized Android touch and keyboard via ADB shell commands
  • vision.mjs: screenshot from page OR device, then Claude/Ollama/OpenAI VLM, then pixel coordinates
  • hybrid-router.mjs: ghost/hybrid/speed mode selection per session, with cdp/os/adb backend selection

All existing V1 humanization modules now accept outputMode: 'os' to route through OS-level APIs instead of CDP.

Pages in this section

  • Installation reproduces the install walkthrough from bible/setup.md
  • Quickstart walks the first 10 minutes from bible/getting-started.md
  • Architecture covers three I/O modes and the hybrid router

Requirements

RequirementMinimumRecommended
OSWindows 10Windows 11
Node.js18.0.020+ (LTS)
ChromeAny recent versionLatest stable
RAM8 GB16 GB
Disk2 GB free10 GB free
TerminalAnyPowerShell 7