Learn AI · Offline
AI that works
even without the internet.
Most AI needs a connection. Some of the most important work cannot have one. Here is how offline AI works and when to use it.
What "offline AI" actually means
An offline / local model runs on your own machine — no cloud request, no data leaving the device. Small models fit on laptops; the terminal's /offline workflow configures 100% offline mode using a local model (Phi-3 and friends), ideal for air-gapped environments and sensitive material.
The honest trade-off: a local model is smaller than the frontier cloud models. It will not match Claude or GPT on the hardest reasoning — but for drafting, search, summarization and private work, it is a strong, private workhorse.
When to go offline
- Air-gapped environments — defense, critical infrastructure, classified-adjacent work.
- Travel — flights, remote sites, patchy connectivity.
- Sensitive documents — anything you cannot send to a third-party API.
- Zero-retention policies — when even ephemeral cloud processing is not allowed.
The hybrid approach most people should use
You do not have to choose one mode forever. A good setup is hybrid:
- Offline mode for the sensitive, the travel, and the routine.
- Online routes (free/pro lanes) when you need frontier reasoning.
- Local search (
/lib) for instant vault lookup at zero cost.
The same terminal switches between all three. Learn how the lanes work in model routing explained, and see the privacy angle in AI privacy & security.