github.comOpen source

AIguessr

An autonomous AI bot that plays GeoGuessr using multi-agent computer vision — no cheats, no data extraction, pure AI reasoning.

Project Overview

AIguessr is a standalone Windows desktop application that connects to your browser and plays GeoGuessr autonomously. Unlike traditional bots that intercept Google Maps API calls to extract exact coordinates, AIguessr uses a multi-agent AI vision pipeline to analyze Street View screenshots — reading signs, checking terrain, identifying road markings, and reasoning about geographic clues, just like a human player would. It connects externally via Chrome DevTools Protocol with zero page injection, and mimics human-like camera panning, zooming, and mouse movements.

Tech Stack
PythonPlaywright (CDP)FastAPIWebSocketsReactViteTailwind CSSClaude / OpenRouter APISQLitePyInstallerInno Setup
Key Features
  • 4 specialist AI agents analyze each round in parallel — Scout, Text Reader, Meta Analyst, Terrain Analyst — then a Synthesizer combines findings
  • Two game modes: Classic (untimed, full analysis) and Duels (1v1 competitive with health tracking)
  • Two play styles: Aggressive (fully autonomous) and Assisted (AI coaching overlay, you place the guess)
  • Supports multiple AI providers: Claude, GPT-4o, Gemini, Grok, DeepSeek via Anthropic API or OpenRouter
  • 120+ instant-ID geographic clues and 57 country-specific intelligence files
  • Real-time web dashboard with game controls, action log, analytics, and dark/light theme
  • Standalone Windows installer — no terminal or Python setup needed

© 2025 Werner's Works. All rights reserved.