What's my local AI?

>_'

skip to content

Get updates

×

We'll let you know when new models or features land. No spam, unsubscribe anytime.

{ if (r.ok) { emailSubmitted = true } else { emailError = true } })<br>.catch(() => { emailError = true })<br>action="https://api.matto.club/waiting-list/signup"<br>method="POST"<br>class="email-form"

Subscribe

Couldn't subscribe you right now — please try again.

Thanks! We'll be in touch.

What's my local AI?

Check what's the best model you can run locally!

This tool needs JavaScript. The detection runs locally in your browser, so there's nothing to upload.

Your machine

reset my specs

We auto-detect what we can from your browser; everything else is a best guess. Tweak any value below to see<br>what else could run.

Hardware

GPU<br>edited

Memory

RAM<br>edited

adjust('ram', -1))"<br>@pointerup="stopRepeat()"<br>@pointerleave="stopRepeat()"<br>@pointercancel="stopRepeat()"<br>@keydown.enter.prevent="adjust('ram', -1)"<br>@keydown.space.prevent="adjust('ram', -1)"

adjust('ram', 1))"<br>@pointerup="stopRepeat()"<br>@pointerleave="stopRepeat()"<br>@pointercancel="stopRepeat()"<br>@keydown.enter.prevent="adjust('ram', 1)"<br>@keydown.space.prevent="adjust('ram', 1)"

VRAM<br>edited

adjust('vram', -1))"<br>@pointerup="stopRepeat()"<br>@pointerleave="stopRepeat()"<br>@pointercancel="stopRepeat()"<br>@keydown.enter.prevent="adjust('vram', -1)"<br>@keydown.space.prevent="adjust('vram', -1)"

adjust('vram', 1))"<br>@pointerup="stopRepeat()"<br>@pointerleave="stopRepeat()"<br>@pointercancel="stopRepeat()"<br>@keydown.enter.prevent="adjust('vram', 1)"<br>@keydown.space.prevent="adjust('vram', 1)"

Platform

CPU<br>edited

adjustCores(-1))"<br>@pointerup="stopRepeat()"<br>@pointerleave="stopRepeat()"<br>@pointercancel="stopRepeat()"<br>@keydown.enter.prevent="adjustCores(-1)"<br>@keydown.space.prevent="adjustCores(-1)"

adjustCores(1))"<br>@pointerup="stopRepeat()"<br>@pointerleave="stopRepeat()"<br>@pointercancel="stopRepeat()"<br>@keydown.enter.prevent="adjustCores(1)"<br>@keydown.space.prevent="adjustCores(1)"

OS<br>edited

WebGPU<br>edited

Your local model

lean

max performance

LM Studio

Ollama

tight fit

even the tiniest model here wants ~2 GB. double-check your numbers above?

Other options

0" x-cloak>

largest first<br>smallest first<br>name a–z

nothing smaller on the list — your pick above is the sweet spot.

0 && visibleAlsoRuns.length === 0"><br>nothing matches — try clearing the search or filters.

0">

tight fit

3"<br>@click="toggleShowAll"<br>x-text="showAll ? 'show top 3' : 'show all ' + alsoRuns.length + ' models'"

What can you do with it?

Local models are private by design — everything stays on your machine. A few things people actually do:

Private drafting

Brainstorm, rewrite, or translate text with no data ever leaving your laptop. Perfect for work that isn't<br>yours to share.

Offline code help

Autocomplete, explain a gnarly snippet, or write a quick script — even on a plane or in a dead zone.

Read your own files

Summarize that 50-page report, pull the action items from a meeting transcript, or chat with your notes —<br>without uploading them anywhere.

Zero-cost experiments

Try open-weight models (reasoning, vision, tool calls) on your own hardware before paying for an API — the<br>model list on this page is your playground.

No throttle, no quota

Run as much as you want. No rate limits, no token metering, no monthly bill — just your electricity.

Learning & tinkering

The fastest way to build a mental model of what different model sizes are good at — swap models and feel<br>the difference.

How detection works

GPU<br>WebGL renderer string

VRAM<br>GPU database lookup, unified memory on Apple, or estimate from RAM

RAM<br>navigator.deviceMemory, cross-checked against the detected chip on Apple Silicon

CPU<br>navigator.hardwareConcurrency

OS<br>user agent parsing

WebGPU<br>navigator.gpu presence