Saturday, 21 November 2026 Press Releases Newsletter RSS
Breaking
WhatsApp Business Account Hacked in Pakistan: Full Recovery Guide Special Technology Zones Authority (STZA) & IT Park Expansion in Pakistan: Islamabad, Karachi, and Lahore Ground Report Pakistan Software Export Board & Industry Consortium Launch 'TechBridge Pakistan 2026' to Connect 500 Startups With Global Markets Global Tech Safety Summit 2026: International Standards for Frontier AI and Data Security Announced WhatsApp Business Account Hacked in Pakistan: Full Recovery Guide Special Technology Zones Authority (STZA) & IT Park Expansion in Pakistan: Islamabad, Karachi, and Lahore Ground Report Pakistan Software Export Board & Industry Consortium Launch 'TechBridge Pakistan 2026' to Connect 500 Startups With Global Markets Global Tech Safety Summit 2026: International Standards for Frontier AI and Data Security Announced
WritePures

Before installing anything, it is worth being clear about what this gets you — and what it does not, because the honest version is more useful than the enthusiastic one.

The four real advantages

Nothing leaves your machine. This is the one that actually changes what work is possible. A lawyer cannot paste a client's file into a hosted chatbot. A clinic cannot upload patient notes. A school cannot send exam scripts to a third party. A model running locally has no such problem, because there is no network request to worry about.

No cost per token. You pay for electricity. Running the same prompt ten thousand times costs the same as running it once, which makes whole categories of batch work viable that would be absurd on a metered API.

It works offline. On a plane, in a village with intermittent connectivity, during the kind of outage that took Microsoft 365 down across seventy countries this month. The model does not know the internet exists.

Nobody can take it away. This is the property the whole tutorial rests on, and it is genuinely new.

Why a downloaded model cannot be recalled

An open-weight model is a file. Once you have downloaded it, you hold a complete, working copy of the thing itself — not a licence to access someone else's copy over a network.

The publisher can remove it from their hub tomorrow. They can change the licence on future versions, be acquired, go bankrupt, or be ordered by a government to stop distributing. None of that reaches the file on your disk. It keeps working exactly as it did the day you downloaded it, indefinitely, with no telemetry and no way to be disabled remotely.

This is the opposite of how hosted AI works, where the model behind an API can be changed, deprecated or repriced without your involvement, and where "the model got worse" is a complaint with no recourse.

The practical implication: if your work depends on a specific model, keep your own copy of it. Weights you rely on should sit on your own storage and in your own backups, not only in a cache that a reinstall clears.

What you should not expect

A 7B or 8B model running on a laptop is not a frontier model and pretending otherwise wastes your time.

What it is good at: summarising documents, drafting and rewriting text, extracting structured data from messy input, classification, translation, answering questions about text you give it, and straightforward coding help.

What it is weaker at: long multi-step reasoning, obscure factual recall, very large contexts, and anything where a subtle error is expensive. Larger local models close some of that gap, and chapter two explains what they cost in hardware.

The useful mental model is not "a free ChatGPT". It is "a capable, private, free assistant for a large share of ordinary text work, available offline".

What you need before chapter two

  • A computer with at least 8 GB of RAM. 16 GB is comfortable.
  • About 10 GB of free disk space to start.
  • A terminal: Terminal on macOS or Linux, PowerShell or Windows Terminal on Windows.
  • A connection, for the downloads. After that, nothing.