What you need
| Windows | Windows 10 or 11, 64‑bit |
|---|---|
| Memory (RAM) | 8 GB is enough to start. More memory makes answers faster. |
| Graphics card | Not needed. An NVIDIA card is used automatically if you have one. |
| Disk space | About 175 MB for Laghu, plus 1 to 20 GB for each model you download |
| Internet | Only to download Laghu and the models you choose |
1. Install Laghu
Download Laghu and open the installer. It installs for you alone, without administrator rights, and there is no account to make. Windows may say it “protected your PC”, because Laghu is new; choose More info, then Run anyway. Why this happens.
When Laghu opens for the first time, it suggests a model that suits your computer’s memory, so you can start straight away.
2. Pick a model for your PC
Laghu has a built-in library of open models from Meta, Mistral AI, the Qwen team, DeepSeek and OpenAI. You see each model’s size and licence before anything downloads. A good place to start:
| Your PC’s memory | For quick chat | For better answers, more slowly |
|---|---|---|
| 8 GB | Qwen2.5 1.5B, Llama 3.2 3B | Qwen2.5 7B, Qwen3 8B, gpt‑oss 20B |
| 16 GB | Qwen2.5 7B, Llama 3.1 8B | Mistral Nemo 12B, Qwen3 14B, gpt‑oss 20B |
| 32 GB or more | Qwen3 8B, Mistral Nemo 12B | Qwen2.5 32B, Qwen3 32B, Mistral Small 24B |
Laghu also runs models that are bigger than your free memory, more slowly: see how to run a large AI model on an 8 GB laptop. Every model in the library is described in Llama, Qwen, Mistral and DeepSeek on Windows.
3. Start chatting
Load the model and type, as you would with any AI assistant. Ask it to explain something, draft an email, rewrite a paragraph or plan a trip. Attach a PDF, Word, PowerPoint, Excel or text file to ask about it, and with a model that can see, a picture too.
For work across many files, such as summarising a folder of PDFs, writing documents or tidying a folder, switch on agent mode. It works in a folder you choose and asks before it acts.
Offline and private
Once a model is downloaded, Laghu works with the internet switched off. Your conversations are kept on your computer, there is no account, and nothing you type is sent to us or to anyone else. Laghu goes online only when you ask it to, to download a model or to search the web, and to check once a day for updates, which you can turn off. Our privacy policy lists exactly what goes online.
With or without a GPU
You don’t need a graphics card. Laghu runs models on your processor, so an ordinary laptop works. If your PC has an NVIDIA graphics card, Laghu uses it automatically and answers faster. Either way, the memory you let Laghu use matters most: more memory, quicker answers.
Questions
Is it free to run an LLM locally with Laghu?
Yes. Laghu is free, with no account and no subscription, and the models in its library are free to download. Each model has its own licence, which Laghu shows before you download it.
What is the best local LLM for Windows?
It depends on your computer and the job. On most laptops, Qwen2.5 7B or Llama 3.1 8B is a good all-rounder; Qwen2.5 1.5B is the quickest; gpt‑oss 20B is the strongest at hard questions, if you can wait.
Can I use models I downloaded with Ollama?
Yes. Laghu finds the models you have downloaded with Ollama, and opens any .gguf file, so you don’t need to download them twice.
Does Laghu work on a Mac or on Linux?
Not yet. Laghu is for Windows 10 and 11, 64‑bit.
How does it compare with ChatGPT?
Online assistants run much larger models and know more; a local LLM keeps everything private and works offline. An honest comparison.