Local AI is private, free and works offline. Tell us what computer you have and we’ll show you which AI models will run smoothly on it, and exactly how to install them.
The check takes about a minute
28 tok/s
Speed
0 bytes
Sent online
$$0
Per month
Dot, our pixel mascot, walks you through exactly what this site does — in 57 seconds.
You download a free app and an AI model. The app runs the model on your own machine. No account, no cloud.
We do the math for your exact computer, so you don’t have to learn it first.
Speed is measured in tokens per second (a token is about ¾ of a word). Tap the speeds to feel the difference. Your computer decides which one you get.
Learn what decides the speedReadable About reading speed. Fine for chat.
Most calculators only check whether a model squeezes into memory. But your computer needs room to breathe too. We rate how a model will feel, from “Technically runs” to “Excellent”.
See why in 3 minutesIt “fits”… but your computer crawls
The model squeezes out your apps. The computer starts swapping to disk and everything, including the AI, slows to a crawl.
Our top picks for chatting and for coding on the machines people ask about most.
Swipe for more. Picks assume a chat app (Open WebUI) and a coding agent (OpenCode) on a medium-sized project.
Prices are re-checked every morning, and each card shows how its price moved since the previous check.
$1,999.99
The fastest consumer GPU for local AI: 32GB of VRAM fits 30B-class models with room to spare. This is the Founders Edition list price — board-partner cards sell for much more.
Checked Oct 20 · Best Buy
View product$999.99
16GB of VRAM is the sweet spot for 14–32B models and fast image generation. Founders Edition list price.
Checked Oct 20 · Best Buy
View product$1,299.99
There is no Founders Edition of the 5070 Ti, so we track a mainstream board-partner card. Its 16GB runs most 14–32B models well.
Checked Oct 23 · Best Buy
View product$6,950
A complete AI computer with 128GB of unified memory: 200B-class models on a desk, with NVIDIA's full CUDA stack built in.
Checked Oct 20 · NVIDIA Marketplace
View product$4,999
The same GB10 chip with half the memory, for models up to about 100B. On sale October 23, only from Acer, ASUS, Dell, Gigabyte, HP and MSI — not from NVIDIA directly.
Checked Oct 20 · NVIDIA
View product$2,499
Apple's base Studio pairs the M5 Max with 36GB of unified memory, and can be configured up to 128GB for the biggest local models.
Checked Oct 20 · Apple Newsroom
View product$899
The cheapest current Mac for local AI: an M6 chip with 16GB of unified memory in a tiny, silent box.
Checked Oct 20 · Apple Newsroom
View product$1,269
A modular mini-PC you can open and upgrade yourself. The base Strix Halo configuration pairs a Ryzen AI Max 385 with 32GB of unified memory.
Checked Oct 20 · Framework configurator (via Pandromeda)
View productAnimated, jargon-free, about 37 minutes in total. Each lesson ends with a quick question so you know it clicked.
Start learningMore tools for when you’re ready
Ratings come from a model of each computer’s memory and speed, calibrated against public benchmarks. How we calculate
