LLM in the Browser

A tiny AI playground running entirely on your machine. The models are small, the results are unpredictable, and that's the whole point. Pick a model, download it once, and see what language models actually do up close.

Fair warning: the 135M model thinks 10+10 equals twelve, you're absolutely right! You have been warned.

This page is a technical demonstration only. Unwrite is not responsible for any content generated by these models. Outputs may be inaccurate, nonsensical, or inappropriate.

100% Private
Browser-Based
Truly Free

Device note: Larger models need more RAM and a faster CPU/GPU. Models over ~1 GB may crash mobile browsers — iOS Safari limits tabs to roughly 1–3 GB even on 8 GB iPhones, and Android Chrome caps around 2–4 GB on flagships. Older or budget devices will struggle with anything above the smallest models.

Welcome! I can chat with you using AI models that run entirely in your browser. Nothing you type ever leaves your device.

Pick a model to get started

Downloads once and caches in your browser. Smaller = faster.

FLAN-T5 Small77MSeq2Seq
2022·90MB·512 ctx·Encoder-decoder, answers not chat
SmolLM2 135M135MDecoderUnfiltered
2024·182MB·2K ctx·Tiny, fast, beautifully unhinged
Florence 2 Base230MVisionVision
2024·275MB·Accepts images·Captioning, OCR, and object detection
SmolLM2 360M360MDecoderUnfiltered
2024·388MB·2K ctx·Better coherence, still wild
Qwen 2.5 0.5B500MDecoder
2024·790MB·4K ctx·Good quality, follows instructions well
Qwen 3.5 0.8B0.8BDecoderPreview
2025·571MB·33K ctx·Needs Transformers.js 4.0.0-next.9+
Gemma 3 1B1BDecoder
2025·763MB·33K ctx·Google's compact instruction model
Llama 3.2 1B1.24BDecoder128K Context
2024·1.1GB·131K ctx·Meta's latest, massive context window
Qwen 2.5 1.5B1.5BDecoder
2024·1.2GB·33K ctx·Best quality-to-size ratio
Phi 3.5 Mini3.8BDecoderLargest
2024·2.3GB·131K ctx·Biggest available, best output quality
Janus Pro 1B1BMultimodalImage Gen + Vision
2025·2.1GB·Accepts images·Generates images and understands them
Messages are processed locally and never leave your browser.

How does in-browser AI work?

Everything on this page runs locally in your browser. These panels explain the technology behind it.