Smart suggestions should not cost you your privacy.
Right now, every song recommendation, every search result, every map route is powered by a copy of your behaviour that lives on someone else's server. You never agreed to that trade — you just clicked "Accept." This page explains the problem, the shift that makes a solution possible, and a proposal for what comes next.
The deal you never made
Every time you ask a voice assistant for the weather, search for a restaurant, save a song, or request a route, a small piece of your life is recorded. Not as a favour to you — as fuel for a machine that turns behaviour into profit.
Your searches reveal what you are curious about. Your music reveals your mood. Your map history reveals where you live, where you work, and who you visit. Your prompts to AI assistants reveal how you think. Individually, each data point seems harmless. Together, they form a portrait more detailed than anything you would willingly hand a stranger.
The companies that collect this data are not breaking any laws. The terms of service you accepted — the ones nobody reads — granted permission. But permission given without understanding is not informed consent. It is a loophole dressed as a checkbox.
Once your data enters a cloud model, it cannot be reliably un-learned. Deletion removes the record, but the model's behaviour has already absorbed its influence. The data is gone; the ghost of it remains.
Where your data actually goes
Understanding the problem means seeing the pipeline. Here is what happens every time you interact with a cloud-based AI service:
You act
A search, a song, a route, a prompt. Something small and ordinary.
It travels
Your request leaves your device and lands on a server you do not control, in a country you may not know.
It is stored
Logged, timestamped, tagged with your identity or device fingerprint, and added to your behavioural profile.
It is used
To train models, target advertisements, refine recommendation engines, or sold to third parties as "anonymised" data — which is rarely as anonymous as claimed.
You get a suggestion
A better playlist. A faster route. A smarter reply. The product feels free because you are the product.
This pipeline has operated for over a decade, and most people have accepted it as the cost of convenience. But it is not the only way.
The shift that changes everything
For years, the argument for cloud AI was simple: the models are too large and too expensive to run anywhere except a data centre. That argument is collapsing.
Hardware caught up
A mid-range desktop with 16 GB of RAM can now run a quantised 7-billion parameter model. That is enough for meaningful personalisation — music taste, search habits, writing style — without ever touching the cloud.
Open models exist
Projects like Llama, Mistral, and Gemma have released capable models under open licences. You can download them, run them on your own hardware, and owe nothing to anyone.
Tools are ready
Software like Ollama, LM Studio, and llama.cpp lets anyone run a local model with a single command. What required a PhD and a rack server five years ago now runs on a laptop.
The pieces are in place. What is missing is not technology — it is the idea that you should keep your data local. That personalisation is a right, not a subscription.
The proposal: a local preference engine
Imagine a piece of software — a one-time download, no subscription — that sits on your own machine and connects to a locally running language model. It does one job: it learns your preferences and gives you suggestions, without your data ever leaving your home.
The core principle: your data should be processed where it was created — on your device, under your control. Personalisation does not require surveillance.
What this means for you
Cloud model (today)
- Your data lives on corporate servers
- Terms of service grant broad usage rights
- Deletion requests are difficult to verify
- Your profile can be sold, leaked, or subpoenaed
- You pay with data — monthly, forever
Local model (the shift)
- Your data stays on your machine
- No terms of service — you own every byte
- Deletion means pressing Delete
- No one can access what does not leave your network
- You pay once — with hardware you already own
This is not about rejecting AI. It is about relocating it. The intelligence stays; the surveillance goes.
A vision worth building toward
We are not proposing that everyone abandon cloud services tomorrow. We are proposing that the option should exist — and that it is closer than most people realise.
A world where your phone's AI assistant runs on a model you downloaded, trained on preferences you curated, stored on a drive you can unplug. Where "personalised" does not mean "profiled." Where the smartest suggestions come from a machine that works for you — not one that reports on you.
The hardware is ready. The models are ready. The missing piece is awareness — knowing that you have a choice, and that the choice matters.
Your data is not trivial. It is a map of who you are — your curiosities, your routines, your vulnerabilities. It deserves the same protection you give your home. The first step is deciding it belongs to you.
Keep reading
See how data risks fit into the bigger picture in Risks & Responsibility, or explore real projects changing the status quo in AI for Good.