ghost

A local AI that runs entirely in your browser. No install. No API key. No GPU. No data leaves your device.

try it

ghost.zaydmulani09.github.io

Open the link. Pick a model. Chat. That's it.

how it works

ghost runs a quantized LLM entirely in WebAssembly inside your browser tab.

• Inference: wllama — WebAssembly binding for llama.cpp
• Models: GGUF quantized models loaded from HuggingFace
• Caching: Models download once and cache in your browser's Cache API — works offline after first load
• RAG: Drag a .txt or .pdf onto the chat — ghost embeds and retrieves relevant chunks locally
• Voice: Web Speech API — speak your message, no mic data leaves your device

models

Model	Size	Speed	Best for
Qwen2.5 1.5B	1.1 GB	~3 tok/s	General use, coding
Qwen2 0.5B	400 MB	~8 tok/s	Quick answers, low RAM
TinyLlama 1.1B	670 MB	~5 tok/s	Lightweight devices

features

• Fully offline after first model download
• Model picker — choose speed vs quality
• Markdown rendering — code blocks, lists, bold, italic
• RAG — chat with your own documents (.txt, .pdf)
• Voice input — mic button, auto-sends on silence
• Multi-turn memory — remembers conversation context
• Copy button on every response
• PWA installable — add to home screen

stack

• Pure vanilla JS ES modules — no build step, no npm
• wllama 2.3.7 (MIT) for WASM inference
• PDF.js 4.0.379 for PDF text extraction
• Web Speech API for voice
• Single HTML file + service worker

local dev

node server.js
# open http://localhost:8000/ghost.html

Requires Node.js. The server sends COOP/COEP headers required for multi-thread WASM.

license

MIT