Why Ngaro exists
Every file you paste into a cloud chatbot leaves your building. Your client records, your contracts, your margins, your case notes, your payroll, all of it goes to a server you don't control, under terms you didn't write, belonging to a company that may be sold, subpoenaed or breached.
Most people know this and paste anyway, because the alternative has been to go without. The tools that keep your data at home have, until recently, required a command line, a config file and a working knowledge of how models are served. That is a real barrier, and it has quietly meant that privacy is a thing you can have if you happen to be technical.
Ngaro exists to remove that trade-off. You describe what you want in plain English, you feed it your documents, and you install one app, the kind you double-click. From there your documents, the index built out of them and every conversation you have stay on your disk. That holds whether the model runs on your own machine or on ours: what you end up with is not a folder on somebody else's server, it is a program on your computer that happens to know your material.
The name, and the owl
Ngaro is te reo Māori for hidden, or unseen. It is the whole product in one word: the thing you build stays out of sight, on your machine, unavailable to anyone you haven't handed it to.
The mark is an owl (ruru, the morepork), which is the obvious companion to that idea. It sees in the dark and it is not seen. It is also, unavoidably, a bird that lives here and not somewhere else, which felt right for a product built in Aotearoa about keeping things close to home.
Who builds it
Ngaro is built by Andre Cronin-Nwofor and Harry Wood, two university students in Auckland.
We think everyone should get the useful half of AI, and that wanting to keep your own documents to yourself should not be the thing that rules you out. That is the whole reason Ngaro exists.
If you would rather ask us something directly before trusting us with anything, ngaro.help@gmail.com reaches both of us.
What we believe
Architecture beats assurance. A promise to handle your data carefully asks you to trust our staff, our contracts, our breach response and whoever owns us in five years. A design in which we never receive your data asks you to trust one checkable thing. We would rather build the second kind of product, even where it costs us the conveniences of the first.
Open weights beat black boxes. The models we run are the Qwen3 family, published openly, and anyone can download, inspect and run them independently of us. They run through Ollama, which is open source too. So the part of this product that actually produces your answers is a part you can check against the public release rather than take on faith.
We do not publish the rest of our engineering, and we would rather say that plainly than let you assume. What decides your privacy is not which libraries we picked; it is where your documents sit, and the security page answers that in full.
You should be able to leave. A self-hosted Ngaro AI runs on your own machine, on a model already sitting on your disk. If we shut down tomorrow it keeps working, free of charge, permanently. That is a commitment rather than a hope, and we are aware it makes us easier to abandon. We think a product that holds you by lock-in rather than by being good is not worth building.
Honesty about limits is part of the product. For general-purpose reasoning the largest cloud models still have an edge over what runs on your laptop, and we say so on the front page rather than hoping you don't notice. The AI can be wrong. On the hosted tiers your question and the passages it pulls from your documents do travel to our compute to be answered, even though nothing is kept there. Telling you these things costs us some sales and buys the only thing that matters more.
What we will and won't do
- We will never train a model on your documents or your conversations. We do not retrain or fine-tune models at all.
- We will never sell, rent or trade your information.
- We will never use your content to improve, benchmark or demonstrate our product.
- We will not publish a testimonial, a client logo or a rating until it is real. There are none anywhere on this site, because we have no customers yet.
- We will never store your documents or your conversations. On every tier they stay on your own machine.
- If we ever shut Ngaro down, self-hosted AIs keep working, free of charge, permanently. If that ever has to change, we will say so loudly and well in advance.
Where we are
Ngaro is built in Tāmaki Makaurau, Auckland, in Aotearoa New Zealand. We serve New Zealand and Australia, and we price in New Zealand dollars.
The company is Ngaro AI Limited, NZBN 9429052572418, registered in New Zealand. The same name and number are on the terms of service and the privacy policy, because a company you are trusting with a decision like this should be one you can look up.
That matters more than a flag on a website usually does. New Zealand businesses handling personal information sit under the Privacy Act 2020, and putting client data into an overseas cloud tool is a disclosure with real obligations attached. Being here means we understand the question people are actually asking when they ask whether this is safe.
Where we're up to
Ngaro is pre-launch. The product described on this site is what we are building and what we intend to ship; it is not yet in the hands of paying customers.
We are saying that here, in plain words, because the alternative, writing in the present tense and letting you assume, is the small dishonesty that most startup websites commit, and it sits particularly badly on a company whose entire pitch is that we tell you the truth about where your data goes.
Talk to us
If you are wondering whether your documents suit this, which tier makes sense, or whether the whole idea is right for you, just ask. We would rather tell you honestly that Ngaro is the wrong fit than sell you something that disappoints you.
ngaro.help@gmail.com, or use the contact form.