Download a local LLM for Windows — free, no account
A local LLM is a model file you download once and run on your own hardware: no chat service in the middle, no account, no meter running. This page is the practical route: which model to take, which quant fits your RAM, and how the whole thing stays updated. Strata does the plumbing — you just pick the model.
Download Strata 0.2.17, extract the zip and run the exe. Allow the administrator prompt.
Setup downloads the app and the engine that matches your hardware: NVIDIA CUDA, AMD HIP tuned per generation, Intel SYCL, Apple Metal, or a plain CPU engine.
Open the app and pick a model from the picker — that is the local LLM download itself.
Step two: pick your model and quant
The recommended model is Qwen3.8-Flash-Next, published by several providers in different sizes. A quant is a compressed build of the same model: more compression means less RAM and slightly less polish. Rule of thumb: smaller answers faster, bigger answers better.
For programming, take the pinned Coder build of the same model — it is tuned for code. The local AI guide covers the privacy and offline side; the local API page covers plugging the model into your tools.
What it looks like
Your local LLM at work: tokens per second, memory use, request slots.
Keeping it current
strata update — or the app's own updater — refreshes the app, the Terminal and the engines while keeping your models, chats and settings. Models themselves come from the public providers and the update feed; a small background program called Control handles engine installs.
FAQ
Which local LLM should I download?
Start with the recommended Qwen3.8-Flash-Next. It comes in several quant sizes: smaller needs less memory and answers faster, bigger answers better. For programming there is a pinned Coder build.
What is a quant, and which one fits my PC?
A compressed build of the same model. 8–16 GB of RAM fits small quants like Q2_0 or IQ2_XS, 16–32 GB fits the middle sizes, and 64 GB lets you keep 47 GiB of experts in RAM.
Does a local LLM need the internet?
Only to download the app, engine and model the first time, and to update later. After that every answer is computed on your hardware with no connection at all.
How do I keep a local LLM updated?
Run strata update or let the app do it. Updates keep your models, chats and settings; engines are downloaded per platform by Control.
Is downloading a local LLM free?
Yes. Strata is free and open source (MIT), and the models are published by their providers at no cost. No subscription, no API key, no per-token bill.
Where do I download a local LLM safely?
From this site or the official GitHub releases page. Avoid re-uploaded builds on forums and file sharing sites.
Only trust this page. Fake AI tool builds float around forums and file sharing sites. The real build matches what the maintainers publish on GitHub.