Local-first article summaries

Read it once. Remember it forever.

ArticleLens distills any article into the key points — so you can recall what mattered later, without hunting back through the page.

Add to Chrome GitHub

webgpu-weekly.dev/native-class-compute

How WebGPU brings native-class compute to the browser

WebGPU Weekly · 9 min read

For most of the web's history the GPU in every laptop was off-limits to web apps except through WebGL — an API for drawing triangles, not crunching tensors.

WebGPU changes the contract. It exposes compute shaders, storage buffers and explicit pipelines through an API that maps closely to Vulkan, Metal and D3D12.

The consequence: workloads that used to demand a native app — from image processing to LLM inference — now run inside a browser tab at a fraction of the old penalty.

Quantization closes most of the rest. A 3-billion-parameter model in q4f16 fits comfortably in consumer VRAM budgets, and…

ArticleLens Llama 3.2 3B · local

Reading article…

WebGPU: native-class compute in the browser

WebGPU exposes the machine’s GPU through a modern, explicit API — making in-browser LLM inference practical, private and free.

→ Explicit pipelines — compute shaders run models without leaving the tab.
→ Quantized weights — q4f16 models fit consumer VRAM budgets.
→ Private by design — the article never touches a server.

12.4s 342 tokens $0.00

Privacy

Private by architecture — not by promise.

The default model downloads once and runs on your GPU. No server, no account, no telemetry — the page you read stays yours.

network requests with your text

100%

of inference on your own GPU

one-time ~2 GB download, then cached

optional cloud providers

How it works

From article to summary in one click.

01
Open the side panel

Click the ArticleLens icon on any article. The summary lives next to the page, pinned to its tab.
02
Choose your model

Run locally on-device via WebGPU, or bring your own cloud key — you pick per summary.
03
Summarize

Title, TL;DR and key points. Long reads are summarized chunk by chunk with live progress.
04
Export it

Download any summary as a clean .md file.

Models

Choose your model.

Three ways to run it — all of them yours.

On-device

Runs in your browser

Llama 3.2, SmolLM3 and Phi over WebGPU. Download once, summarize forever.

$0 — no account, no key

Free cloud

OpenRouter free models

Gemma, GPT-OSS and more at $0 per token with a free API key.

$0 — free key required

Your key

OpenAI · Anthropic

Bring your own key for frontier models, streamed straight into the panel.

~$0.001 — per summary, billed by your provider

Open source

Read the code that reads your articles — every claim here is verifiable.

View on GitHub

Summaries that never leave your machine.

Free, local-first, and open source.