79% of ChatGPT level
About 2 GB
Runs in:
Fits on:
· needs ~2 GB · Free to download · License: Llama 3.2 Community License
Run AI on a Raspberry Pi or old PC
Small models for weak hardware.
The tools below run on your computer. Most of them are free, and the local work stays on your machine.
Works offline · Free options
Your computer
Describe your computer
Pick the closest line. Office PC and Mini PC are estimates. Linux and a home server stay in the list. We guess the specs, then hide models that would not fit. The guess stays cautious.
6 results
Most tools here are free.
Writing Audio & Music Not on the chat board
A tiny voice
Runs in:Fits on:· needs ~0.1 GB · Free to download · License: GPL-3.0
Audio & Music The quick Whisper. More mistakes on names and accents, and it runs almost anywhere.
Runs in:Fits on:· needs ~0.5 GB · Free to download · License: MIT
whisper.cpp
The small engine behind many transcription apps, including on Windows and Linux.
Installer only · Some setup · Free
HONEST COMPARISON
Local or cloud: Run AI on a Raspberry Pi or old PC?
Reviewed Oct 2026
- Privacy: Local winsThe device can work with no network
- Cost: Local winsNo token bill
- Quality: Cloud winsAn API is simply a stronger model
- Ease: Cloud winsAn API key is less tuning than a local runtime
Local AI
A small local model is the point of a Pi or an old PC. A cloud API needs the network and a bill.
- It runs with the cable unplugged
- Llama 3.2 3B and Whisper small fit the brief
- Home Assistant stays on the deviceThe usual always-on house hub runs on your hardware, not in a cloud account.
Where it falls short
- The answers are smaller and slower
- A Pi is a poor place for images or video
Cloud & apps
A cloud API is smarter if the little computer is allowed to phone home.
- CChatGPTCloud AIFreemium
A browser chat when the little computer is allowed to send the prompt away.
Best for: Quick questions
Prompts are uploaded
- OOpenAI APICloud APIUsage-basedPopular choice
A small device can call a much stronger model, and it is no longer offline.
Best for: Stronger answers
Prompts leave the device
Our take
If the goal is a small offline computer, stay local. Call an API only when the answer matters more than the privacy and the network can be up.
Choose local if…
- It has to work offline
- You want no API bill
Choose cloud if…
- You need a stronger answer
- The network is reliable
How to do it in 3 steps
- Open the model page and pick any app under Runs in, for example AnythingLLM, GPT4All or Jan. Each app page links to that maker's own guide.
- Open the Llama 3.2 3B page and check the traffic light for a few kinds of computer.
- Follow that official guide. This site does not add its own install steps.
Questions
- Does this upload my files?
- The local tools on this page don't upload your files. The cloud options in the comparison do.
- What does it cost?
- Most options are free. A few apps are a one-time purchase, and a few models need a computer with more memory.
- What if it will not fit?
- Use Check my computer, or turn on the lighter memory filters. A red light on the model page means the file is too big.
Not into local AI? Try these instead
Ordinary apps for code & tech. These are normal links to the maker's site.