Skip to content

vbomfim/openworld

Repository files navigation

OpenWorld

CI CodeQL License WebGPU


OpenWorld is a browser-based conversational world simulator powered by a local AI model running entirely in your browser via WebGPU. No backend, no cloud — everything stays on your device.

How It Works

  1. Open the app — select and download an AI model (cached after first download)
  2. Describe a scene — "A coffee shop in downtown Seattle on a rainy afternoon"
  3. Meet someone — the AI generates a unique person with a full backstory
  4. Chat — have a natural conversation. They remember you next time.

Features

  • 🧠 100% Local — AI runs in your browser via WebGPU. No data leaves your device.
  • 🎭 Rich Personas — AI generates people with coherent backstories, preferences, habits, and personalities.
  • 💾 Persistent Memory — Personas remember past conversations.
  • 🌍 Any Setting — Modern coffee shops, 1920s Paris, medieval taverns, space stations.
  • 🎬 Living Scenes — The AI narrator introduces events to make scenes feel alive.

Requirements

  • Browser: Chrome 113+, Edge 113+, or Safari with WebGPU enabled
  • GPU: Modern GPU (integrated Intel GPUs work but slower)
  • Storage: ~1-4 GB for the AI model (cached in browser storage)

Development

npm install
npm run dev

Tech Stack

  • Svelte 5 + Vite + TypeScript
  • WebLLM (@mlc-ai/web-llm) — in-browser LLM inference via WebGPU
  • Tailwind CSS — styling
  • IndexedDB (via idb) — local persistence for personas and memories

About

Meet people. Have conversations. All in your browser. A conversational world simulator powered by in-browser AI via WebGPU.

Resources

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors