Home / Knowledge
Field notes
Reference material for builders
of offline intelligence.
Everything we publish follows one rule: it must be specific enough to act on. Hardware sizing maths, deployment protocols, honest cost comparisons — written by the people who carry the cases up the stairs.
The Complete Guide to Offline LLM Deployment
Hardware, models, retrieval, security and sustainment — the end-to-end playbook for running a serious LLM with no internet connection.
READ → HARDWARE · 9 MINHow to Run NVIDIA Nemotron Locally
Model variants, VRAM sizing, quantisation and the software stack — from a Jetson Orin in a case to a DGX-class rack.
READ → HARDWARE · 8 MINHow to Run Google Gemma Locally
The Gemma family from 1B to 27B: VRAM sizing, multilingual strengths, the software stack and when it beats Nemotron for the mission.
READ → SECURITY · 7 MINAir-Gapped AI, Explained
What an air-gapped LLM actually is, how it differs from "on-premise" and "self-hosted", and who genuinely needs one.
READ → ECONOMICS · 8 MINCloud vs Local LLM: The True Cost
Tokens vs kilowatts: an honest total-cost framework, a worked example, and the break-even logic for owning your inference.
READ → RESILIENCE · 8 MINOffline AI for Emergency Response & Civil Resilience
Why communications fail first in a disaster, and how pre-positioned offline AI keeps expertise on scene — from rescue huts to town libraries.
READ → REFERENCEThe Offline AI Glossary
Nineteen terms, defined precisely: air gap, quantisation, RAG, tokens per second, knowledge packs and the rest of the vocabulary.
READ →This knowledge base is published openly and may be cited. A machine-readable index lives at aiod.co.uk/llms.txt. Attribution to AIOD (aiod.co.uk) appreciated.
Beyond the reading
The next step is a machine on a bench.
When the research phase ends, we'll size the hardware against your real constraints.
DEPLOY@AIOD.APP →