Most AI tools work by sending your input to a server — a server you don't control, operated by a company whose data practices you're trusting. For personal tasks that's often fine. For work that involves confidential information, client data, or anything sensitive, the question of where your words go starts to matter.
Running AI on your own device is the answer to that question. Nothing leaves the machine. Nothing is stored remotely. The capability is entirely yours.
Who this matters for most
You don't have to work in a regulated industry for data privacy to be a legitimate concern. It comes up in situations that are easy to imagine:
- Legal and professional work. Client matters, case notes, draft advice — material that has confidentiality obligations attached.
- Healthcare. Patient information, clinical notes, any work governed by data-protection rules.
- Finance. Deal information, portfolio data, anything that would be sensitive if it appeared in a third-party system.
- Personal and creative work. Journals, correspondence, drafts that you simply don't want anywhere other than your own machine.
The common thread is: information where the cost of it going somewhere unexpected is too high to accept.
What on-device AI actually means
When a model runs locally, the processing happens on your hardware. Your input — the document you're summarising, the question you're asking, the text you're editing — is never transmitted anywhere. There is no server receiving it, no company logging it, no account it could be tied to.
This is different from a privacy policy that says "we don't sell your data." It's a stronger guarantee, because the architecture makes the risk impossible rather than just prohibited.
The askFinz extension brings this capability to the work you're already doing — reading, drafting, researching — and keeps it private by design.
The practical trade-off
On-device models are capable, and they're improving quickly. They're well-suited to summarisation, drafting, editing, classification, and most language tasks that knowledge workers do every day. For tasks that genuinely require the most capable frontier models, those are available through the cloud workspaces — but for sensitive work, the local option is there, and for many tasks it's all you need.
Further reading
- Keeping work private when you use AI — a broader look at the privacy questions that come with using AI in professional contexts.
- One AI workspace instead of ten browser tabs — how on-device capability fits into a broader AI workspace.
- The Electronic Frontier Foundation publishes accessible writing on data minimisation and what "privacy by design" means in practice.