browser-llm

Browser LLM inference on hand-written WGSL kernels — up to a 35B sparse MoE no compiler stack ships. No TVM, no WebLLM runtime. Measured; every number in BENCH.md.

AI in your browser that supports MCP Servers, no network call, no costs, 100% private

Free, private AI in your browser

On-device RAG app (WebLLM + Transformers.js + pdf.js). Drop a PDF, ask questions, get cited answers. Runs fully in the browser; data stays local.

A zero-dependency Node.js HTTPS-to-HTTP proxy that lets browser-based AI tools (Agile V Studio, Framewrk Studio) talk to local LLM backends like Ollama without mixed-content errors. Includes self-signed TLS, CORS, token tracking dashboard, Docker support, and OpenAI/Anthropic-compatible passthrough.

High-performance, zero-dependency JS/TS Language Model & NLP suite for Node.js and browsers featuring text generation, self-attention heatmaps, RAG fact injection, and webpage autocomplete.

Run small LLMs directly in your web browser, no cloud computing needed.

Scaffold for browser-local LLM chat apps. Runs Gemma 4 entirely in the browser via MediaPipe + WebGPU — no server, no cloud, no tokens. Mobile-first PWA, Atomic Design, i18n (DE/EN), multi-conversation. Fork it. Own it. Ship it.

Offline-first AI assistant (browser LLM + RAG) for newcomers & disconnected regions. No API keys, works in airplanemode.

🤖 Free LLM chat + 👾 AI-powered arcade game, running 100% in your browser — offline, private, no server, no API key (WebLLM/WebGPU + wllama/WASM fallback)

Starling - Lightning Fast Autofill Extension

Terminal-style portfolio website that runs a local LLM in your browser. Type `ask` to chat with an on-device LLM (Gemini Nano / WebLLM via WebGPU) — no backend, no API keys, your messages stay local. Built with React + Vite.

Pushing fused WebGPU transformer kernels to max model size — int4, tiled FFN, Phi-3-mini 3.6B in Chrome

Browser-native LLM chat powered by Transformers.js, WebAssembly and WebGPU.

在浏览器中用 WebGPU 全本地运行 DeepSeek-R1-Distill-Qwen-1.5B 推理模型，无需服务器、数据不出浏览器。基于 Transformers.js + ONNX Runtime Web，支持流式输出、思考过程可视化、Markdown 与数学公式渲染，代码含详尽中文注释，适合学习 WebGPU / Web Worker 工程实践。

Privacy-first document editor with local in-browser LLM inference via HuggingFace Transformers.js, WebGPU, and Tiptap.