Why 8 GB usually isn’t enough
AI models are big. A 7-billion-parameter model is about 4.4 GB on its own, and a 20-billion-parameter one about 12 GB. Most ways of running AI at home need the whole model to fit into the memory your computer has free, and on an 8 GB laptop, Windows and your browser have usually taken most of it before you start.
Laghu works within the amount of memory you give it, so the model does not have to fit. A model bigger than your free memory still runs; giving Laghu more memory makes it faster. No graphics card is needed.
1. Install Laghu
Download Laghu and run the installer. It needs about 175 MB, works on Windows 10 and 11 (64‑bit), and doesn’t ask for administrator rights. There’s no account to make.
Windows may say it “protected your PC”, because it doesn’t know us yet. Choose More info, then Run anyway. Why this happens.
2. Choose a model that suits your laptop
Open Models and pick one from the built-in library. You see each model’s size, and its licence, before anything downloads. If you already use Ollama, Laghu finds the models you have downloaded there.
- For quick answers: Qwen2.5 1.5B or Llama 3.2 3B. Small, and fast enough for back-and-forth chat on almost any laptop.
- For better writing and reasoning: Qwen2.5 7B, Qwen3 8B or Llama 3.1 8B. Noticeably smarter, and slower.
- For the hardest questions, when you can wait: gpt‑oss 20B. It thinks before it answers, and is worth the wait for maths, logic and planning.
Start with a small model to see how your laptop feels, then try a bigger one.
3. Set how much memory it may use
Leave the memory setting on Automatic and Laghu chooses a good amount each time a model starts, never more than your computer has free. If you want quicker answers and can spare memory, slide it up; if your other apps feel cramped, slide it down. The setting sits in the message box, so you can change it any time.
What speed to expect
We measured these on an everyday laptop: an AMD Ryzen 5 5500U with 7.4 GB of memory, of which about 1.4 GB was free. Speed is in tokens per second; a token is a word or part of one, and a 300-word answer is about 400 tokens.
| Model | Size | Memory given | Speed | A 300-word answer |
|---|---|---|---|---|
| Qwen2.5 1.5B | 0.9 GB | 0.9 GB | 7.0 tokens/s | about 1 minute |
| Qwen2.5 3B | 1.8 GB | 1.0 GB | 2.4 tokens/s | about 3 minutes |
| Qwen2.5 7B | 4.4 GB | 0.8 GB | 0.55 tokens/s | about 12 minutes |
| gpt‑oss 20B | 12.1 GB | 1.0 GB | 0.9 tokens/s | about 7 minutes, plus its thinking |
So on a modest laptop, small models are for chatting and big ones are for jobs you can leave running: a careful letter, a summary of a long document, a hard question. On a computer with more memory, the same models answer much sooner.
How to make it faster
- Free up memory first. Close browser tabs and large apps before you start a model. On Automatic, Laghu then takes more, and answers sooner.
- Plug in. Laptops slow themselves down on battery.
- Match the model to the task. A small model is plenty for rewording an email; save the big ones for questions that need them.
- Keep the model loaded. The first answer after a model starts takes longest; the ones after it come quicker.
- Use a graphics card if you have one. Laghu uses an NVIDIA card automatically. It isn’t needed, but it helps.
- Add memory. If your laptop can take more RAM, it’s the biggest speed-up there is.
Questions
Will it slow down my other apps?
It uses only the memory you allow, so your other apps keep their room. While it writes, your processor is busy, so heavy work such as video editing is best done at another time.
Does it need the internet?
Only to download Laghu and the models you choose. After that it works offline.
Is what I type private?
Yes. Everything runs on your laptop, and nothing you type is sent to us or to anyone else. Our privacy policy says exactly what goes online, and when.
Can it read my documents?
Yes. Attach a PDF, Word, PowerPoint, Excel or text file to a message and ask about it, or switch on agent mode to work through a whole folder of them. How it works.