AI tools explained simply
Local AI: When the model stays in-house.
AI that runs entirely on your own machine – without a single word leaving the premises: that is the promise of local AI models, and the promise holds. The more interesting question is whether you need it. Here is the honest assessment: what works well locally today, what hardware and upkeep cost – and for whom the cloud with a proper contract remains the more sensible route.
- Explained without hype
- Hardware reality
- Honest cloud comparison
- Last updated: October 2026
What “local AI” means.
Normally, AI models run in the providers’ data centres – your inputs travel there and the answer comes back. Local models turn this around: a freely available language model is installed on your own computer or server and works there, even entirely without internet. Tools such as Ollama or LM Studio have made installation as easy as that of a normal program.
The price: local models are smaller than the big cloud models – and noticeably less capable at complex tasks. For many everyday tasks, they are still enough.
What works well locally today – and what does not.
- Good: summarising, rephrasing and classifying texts, simple drafts, translations for in-house use
- Good: processing genuinely sensitive documents that must not reach any service provider
- Good: high-volume automations where per-request cloud costs would add up
- Weak: complex research, demanding texts, multi-step reasoning – the big cloud models are clearly ahead here
- Weak: anything needing current world knowledge – a local model only knows its training cut-off
Rule of thumb: the more sensitive the data and the more repeatable the task, the more local pays off. The more demanding the thinking, the more cloud.
The hardware question, answered honestly.
Small models run on a current business notebook with 16 gigabytes of RAM – they get noticeably faster with a dedicated graphics card or on current Apple machines with plenty of unified memory. Serving several employees requires a small server with a GPU – and someone to look after updates, models and security.
That is exactly where the hidden price sits: local AI has no subscription fee, but costs time and ownership. Without a person who takes care of it, the installation goes stale faster than it creates value.
Or rather cloud with a proper contract?
For most small businesses the pragmatic route is a paid account with a serious cloud provider, with a data processing agreement and the assurance that inputs are not used for training. That covers the data protection requirements of daily business – at the full capability of the large models. What to look for is covered in the guide ChatGPT and data privacy.
Local AI then is not either-or, but the complement for the sensitive remainder – and a building block that can find its place in automations with tools like n8n.
Frequently asked questions.
Is local AI really completely offline?
Yes – after the one-time model download, everything runs on your hardware, even without an internet connection. Nothing leaves the machine unless you yourself connect the tool to online services.
Which models are suitable to start with?
Freely available open models in small and medium sizes, installable in a few clicks via tools like Ollama. The tool shows which variant fits your hardware during setup – start, try, compare.
Does a local model replace ChatGPT or Claude?
For simple, recurring tasks often yes; for demanding reasoning, current information and complex texts, no. The big cloud models remain clearly superior there – realistically, the two coexist.
Is local AI automatically GDPR-compliant?
It solves the transfer problem, because no data flows to third parties – a real advantage. The remaining duties stay: access protection on your own system, deletion concepts and sensible handling of the results.
Prefer not to handle this yourself?
Tell me briefly what it is about – you get an honest assessment and a personal first conversation.
