The beginner — “I just want to try it, no fuss.”
You've heard about local LLMs and you want to see how they run on your own machine. No code, no complicated command line — by the end you'll be chatting with your first model, offline, for free.
By the end of this tour, Ollama will be running quietly in the background, you'll have picked and downloaded your first model, and you'll have a clean web interface to chat with it — all without a single line of code to understand.
Who this tour is for
We'd rather tell you upfront than let you lose half an hour for nothing.
- ✓You barely know what an LLM is, but the topic intrigues you.
- ✓You want to try it without handing your questions to a cloud service.
- ✓You're looking for the shortest path between “I've heard about it” and “it's running on my machine.”
- ✓You have a PC or Mac less than 5 years old with at least 8 GB of RAM.
- ·You want a local coding copilot in your editor: see the developer tour.
- ·You need to query your own documents privately: see the confidential-work tour.
- ·Your machine is from 2015 with 4 GB of RAM: it'll run, but it'll crawl.
The path in 6 steps
Each step links to a real, detailed guide. The order is deliberate — don't skip steps the first time through.
-
1Step 01 · 2 min
Find out what your machine can run
Answer a few quick questions about your GPU, RAM and OS and get a shortlist of models sized correctly for your hardware — no guessing.
You know which model sizes and quantizations will actually run well on your machine.Read the guide → -
2Step 02 · 5 min
Understand what “local” really means
The foundational primer: what a local LLM is, why people bother, and what to realistically expect versus ChatGPT.
You understand the trade-offs before installing anything.Read the guide → -
3Step 03 · 3 min
Install Ollama
The fastest way to get an inference engine running in the background — download, double-click, done.
Ollama is installed and listening on localhost:11434.Read the guide → -
4Step 04 · 3 min
Pick your first model
A ranked, up-to-date shortlist of the best free, open-weight models to start with, so you're not guessing which one to download.
You've picked a model and can pull it with a single ollama run command.Read the guide → -
5Step 05 · 8 min
Get a ChatGPT-like interface
The terminal is austere. Open WebUI gives you a polished chat interface with history, markdown and file uploads — one Docker command.
You're chatting at localhost:3000 with a familiar, browser-based interface.Read the guide → -
6Step 06 · 4 min
See how it really compares to ChatGPT
Honest expectations: where a local 7–8B model holds up against ChatGPT, and where the cloud still wins.
You know exactly when to reach for local and when to still open a browser tab.Read the guide →
The recommended models
Three picks for this tour, lightest to most capable. Click through for the full spec sheet — VRAM, context, benchmarks.
The default starting point. Apache 2.0, solid all-round quality, and a 5 GB Q4 download that fits almost anything.
See the page →Excellent quality-to-size ratio — strong at code and reasoning for a 7B model, and fully multilingual.
See the page →Meta's Llama 3.1 8B — a 131k context window and longer, more structured answers than most 7B models.
See the page →Frequently asked
The real questions readers send in. If yours is missing, reach out via Contribute below.
Does it use a lot of electricity?
During a conversation: yes, your GPU runs close to full tilt (200–400 W depending on the model). Idle between questions: almost nothing. For occasional use it's negligible next to a gaming session.
Are my conversations really private?
Yes, entirely. Once Ollama and a model are installed, you can disconnect from the internet and it keeps working. Nothing leaves your machine — no telemetry, no account, no analytics.
Is it really free?
Yes. Ollama is open source (MIT), and the models above are open-weights with permissive licenses. No subscription, no quota, no “pro” tier — you only pay for electricity.
Why is it worse than ChatGPT?
A 7–8B model has a few billion parameters; GPT-4-class models likely have well over a trillion. For something that fits in 5–6 GB of VRAM, the gap is smaller than you'd think — and a local 70B model closes most of it.
What if I want to try a different model afterward?
ollama run llama3.2, ollama run qwen2.5, ollama run gemma2 — the full library is at ollama.com/library. Each model downloads once, then launches instantly.
The developer — “Local copilot, 100% offline, zero cloud latency.”
A local coding model wired into your editor. Autocomplete, chat, whole-file refactors — without a single line of your code leaving your machine for Microsoft, OpenAI or A…
A question, a typo, a bug?
This tour evolves with every model release. Your feedback is the raw material.