Voice Out
Kokoro 82M
Hexgrad202588 MBTTS
English text-to-speech with 28 voices (American + British accents)
LLM in the Browser
Open-source LLMs running in your browser. Nothing leaves your device. Small models are rough and nothing like ChatGPT; larger ones are more capable but slow to download.
Third-party weights. Unwrite does not host, train, or moderate them. Some are unfiltered; check each model card and licence before use.
100% PrivateBrowser-BasedTruly Free
On-device AI
Pick a model to run in your browser
Downloads once, then caches in your browser. Nothing leaves this device.
Detecting device capabilities...
Unfiltered
SmolLM2 135M
Hugging Face2024182 MBDecoder2K ctx
Tuned with DPO for on-device chat, text rewriting, and function calls
Chat
LaMini Flan-T5 248M
MBZUAI2023260 MBSeq2Seq1K ctx
Fine-tuned on 2.58M LaMini instructions for general instruction following
Chat
LFM2.5 350M
Liquid AI2026280 MBHybrid33K ctx
Designed for tool calling and structured extraction; covers 9 languages
Vision
Florence 2 Base
230MMicrosoft2024320 MBVision
Single model for captioning, object detection, OCR, and phrase grounding
Image input
Unfiltered
SmolLM2 360M
Hugging Face2024388 MBDecoder2K ctx
Stronger at reasoning and instructions while staying fully on-device
Summary
BART Large CNN
406MMeta2022410 MBSeq2Seq1K ctx
Dedicated news summariser fine-tuned on CNN/DailyMail text-summary pairs
Chat
Qwen 3.5 0.8B
Alibaba2025571 MBDecoder33K ctx
Toggles thinking mode on or off; covers 100+ languages and agent tasks
Chat
Gemma 3 1B
Google2025763 MBDecoder33K ctx
Lightweight multilingual chat supporting 140+ languages
Chat
Llama 3.2 1B
1.24BMeta20241.1 GBDecoder4K ctx
Distilled from larger models for fast on-device chat and tool use
Code
Qwen 2.5 Coder 1.5B
Alibaba20241.3 GBDecoder33K ctx
Handles code generation, fixing, and reasoning across 40+ programming languages
Voice In
Granite 4.0 1B Speech
IBM20261.8 GBASR
Speech recognition and translation across 6 languages, ranked #1 on OpenASR
Audio input
Chat
Qwen 2.5 1.5B
Alibaba20241.8 GBDecoder33K ctx
Tuned for coding, math, and reliable structured output like JSON
Chat
Phi 3.5 Mini
3.8BMicrosoft20242.3 GBDecoder131K ctx
Trained on textbook-quality data for strong reasoning and coding
Reasoning
SmolLM3 3B
Hugging Face20252.7 GBDecoder128K ctx
Supports togglable thinking mode and tool calling; fully open weights
Image Gen
Janus Pro 1B
DeepSeek20252.9 GBMultimodal
Generates images from text and answers questions about image content
Image input
Multimodal
Gemma 4 E2B
2.3BGoogle20263.4 GBMultimodal128K ctx
Accepts text, images, and audio; reasons and generates text in 140+ languages
Image input
Model types
- Decoder
- The most common architecture for chat. Generates text left-to-right, one token at a time. Used by GPT, Llama, Qwen, Gemma, Phi, and SmolLM. Good for conversation, creative writing, and general instruction-following.
- Seq2Seq
- Encoder-decoder models that read the full input before generating output. Better for structured tasks like translation, summarisation, and Q&A. FLAN-T5 uses this architecture.
- Hybrid
- Combines convolution and attention layers for efficient on-device inference. LFM2.5 from Liquid AI uses this novel architecture, achieving strong performance at very small sizes.
- Vision
- Processes images as input and produces text descriptions. Florence 2 can caption images, read text via OCR, and detect objects.
- Multimodal
- Handles multiple input or output types. Janus Pro generates and understands images. Gemma 4 E2B accepts text, images, audio, and video.
- TTS (Text-to-Speech)
- Converts written text into natural-sounding audio. Kokoro produces speech across 54 voices and 8 languages.
- ASR (Speech Recognition)
- Converts spoken audio into text. Granite 4.0 is the top-ranked open model on the OpenASR leaderboard.
How does in-browser AI work?
Everything on this page runs locally in your browser. These panels explain the technology behind it.
Thank you to the open-source ecosystem powering this page.
We gratefully acknowledge Hugging Face Hub, Transformers.js, ONNX Runtime Web, the ONNX Community, and model authors/publishers including Hexgrad, Hugging Face, MBZUAI, Liquid AI, Microsoft, Meta, Alibaba, Google. Please review each model card and licence before use.