Interactive

Local model lab

This page runs a real AI model on your computer, inside your browser. Pick a model below, download it once, and chat with it. No account, no subscription, no cloud — and nothing you type ever leaves this page. This is the strongest version of our standing advice: when member data is involved, the safest AI is the one that never phones home.

What can and can't run here

Honest limits first. A browser tab can run small models — roughly half a billion to a few billion parameters, compressed to fit in memory. They handle drafts, rewrites, brainstorming, and summarization well. They are not the right tool for deep research or complex analysis.

Think of it as a ladder with three rungs:

1. This page (free). Small models in your browser. Nothing to install, nothing leaves your machine. Best first step.
2. A desktop app (free). Tools like Ollama or LM Studio run bigger models — 8 to 30 billion parameters — on an ordinary office computer, still fully local. The upgrade kit below walks through setup.
3. A paid subscription. When you need the smartest models or heavy daily use, a team subscription is the honest answer — which is why we've written about budgeting a real AI line item. Keep member PII off this rung unless your vendor contract says otherwise.

Big models with hundreds of billions of parameters can't run on office hardware at all — anyone telling you otherwise is selling something. The ladder above is the whole story: start free and local, spend money only where the work justifies it.

Trusted sources only

Anyone can upload a model to Hugging Face, and anyone does — including strangers with quantized copies of famous models that nobody verified. So this page has a rule: every model here is published by the lab that made it, or by a named partner we trust. No random uploads, ever.

The browser models in the picker are ready-to-run builds compiled and published by the WebLLM team themselves (their mlc-ai organization on Hugging Face), built from the labs' original weights — Alibaba's Qwen, Hugging Face's SmolLM2, Microsoft's Phi. The download happens once; afterwards the model lives in your browser's cache and loads instantly. Your words are never sent anywhere — the download is the only network traffic, and it's the same public file anyone can fetch.

The same rule follows you up the ladder. The desktop route runs on Ollama's curated model library, where the Ollama team packages the labs' official weights. If you ever outgrow a desk-side machine, the chipmakers keep official libraries of their own: NVIDIA NIM and NVIDIA's official nvidia organization on Hugging Face, AMD's official amd organization and Ryzen AI model zoo, and Meta's official meta-llama organization. One honest limitation: Meta's and Google's models require you to accept their license on Hugging Face first, so they aren't one-click loads — which is why the picker sticks to openly licensed models. The upgrade kit covers running gated models as a desktop app, where you supply your own access.

Download the upgrade kit

The lab above is rung one. The kit takes you to rung two: a step-by-step guide to installing Ollama and running larger models (8B and up) on your own machine, plus a prompt pack written for association work — session descriptions, renewal nudges, board summaries — tuned for smaller local models. Free with your email.