NGARO
Private by default 000
A ruru perched on a mossy outcrop at golden hour, backlit by low sun

Your Data
Your Rules
Your AI

Like ChatGPT, but it runs on your own computer
and answers from your own files.

Connect your AI to the apps you already use

Gmail Outlook Google Sheets Notion Slack Xero HubSpot Google Drive Trello Airtable Dropbox Shopify Asana Microsoft Teams

The AI jobs you’ve been avoiding.

You already use AI for the safe things: a social post, tidying up an email. The genuinely useful work sits in files you’d never paste into a chatbot. Client records, contracts, quotes, staff manuals. That’s the work Ngaro is for.

01

Ask your own files

“Which of our quotes last year included scaffolding?” Ngaro reads your invoices, PDFs and spreadsheets and answers from them, then tells you which document it got it from. No more digging through folders for something you know you’ve already answered.

02

A helper your staff can ask instead of you

Point it at your policies, manuals and procedures, then hand it to your team. New starters ask it how the job is done. You stop being the bottleneck for the same question a hundred times.

03

Drafts that sound like you

The quote, the client email, the report, all written from your own past documents, so it comes out in your format and your language instead of generic AI mush. It’s seen how you did it last time.

Summarise a long contract Compare two versions Pull the numbers out of a report Answer client questions

Why Ngaro exists

Every file you paste into a cloud chatbot leaves your building: your client records, your contracts, your margins. Ngaro exists so that never has to happen again.

Not a promise about how we handle your data, but an architecture in which we never receive it. Here is what that actually buys you.

0 documents

reach us, on any tier. Your files, the index built from them and every conversation stay on your disk, so there is nothing here for us to lose.

100% open models

The models that answer you are open: anyone can download them, inspect them and run them without us. They are the Qwen3 family, and they are not ours to switch off.

Web search off by default Flat monthly price Built in Aotearoa Apps connected via API

Yours, forever

If Ngaro ever shut down, self-hosted AIs keep working, free of charge, permanently. The model is already on your disk.

Four steps, no jargon.

You approve every stage, and you always see the whole plan before anything runs. Close the tab halfway through and you’ll pick up exactly where you left off.

Describe the job in your own words. You get back a plan: what it will read, where it will build, the rules it will hold to, and what it will refuse to do. Nothing runs until you approve it, and you can change any part of it first.

Drop in PDFs, Word files, notes or links, or let it research and gather sources for you. Then sign off the complete list, marked by where each item came from, before any of it is built in.

Chat with the real model in your browser, and set the name, greeting and colour the people using it will actually see. Change anything and try again as often as you like: your subscription is one flat monthly price, so rebuilding costs the same as not rebuilding.

Install the Ngaro app on Windows or Mac and sign in. Choose self-hosted and it downloads the model onto your machine; choose hosted and it calls a larger one we run. Either way your documents and every conversation stay on your disk, and the app is where you manage your subscription, your devices and the chats you share between them.

01

Say what you need, see the plan

Describe the job in your own words. You get back a plan: what it will read, where it will build, and the rules it will hold to. Nothing runs until you approve it.

02

Feed it, then check what it knows

Drop in PDFs, Word files, notes or links, or let it research and gather sources for you. Then sign off the complete list before any of it is built in.

03

Preview it, and make it yours

Chat with the real model in your browser, and set the name, greeting and colour the people using it will actually see. Try it as many times as you like: one flat monthly price, whatever you build.

04

Take it home

Install the Ngaro app on Windows or Mac, sign in, and choose where your AI runs. Your documents and your conversations stay on this machine either way.

One key. Everything else plugs in.

Private by default, connected the moment you say so. Your finished AI can hand out a key that speaks the shape every tool already understands, so it talks to your other software without us building a single integration.

The key

Two places it can answer from

Same assistant, same documents, same disk. The only difference is where the model doing the thinking runs.

On your machine http://localhost:8420/v1 Never leaves your network
Model hosted by us https://api.ngaro.net/v1 A bigger model, called from your machine
API key ngr_live_7f3a9b2c4d6e8f01 Yours to issue and destroy

One AI only Rate limited Revoke any time Trains nothing

The reach

Open weights where everything connects

The models that answer you are open weight and independently verifiable, so the part that produces your answers is a part you can check against the public release.

Connecting to apps you know

Gmail Google Sheets Outlook Slack Google Drive HubSpot Xero Notion

Using a simple bridge like

Make Zapier n8n Power Automate

Connect your AI to thousands of apps. Your spreadsheets, your inbox, your CRM, your helpdesk.

We’re not affiliated with any of the tools named here. They speak HTTP, and so do we. Nothing on this list needs our permission, or theirs.

Join the waitlist.

The private preview is open, and we’re letting people in a handful at a time. Put your name down and you’ll be among the first to build an AI that never leaves your machine, plus you’ll hear the launch date before anyone else.

Based in Tāmaki Makaurau, Auckland · Aotearoa New Zealand
Serving New Zealand & Australia

By submitting this form I consent to Ngaro collecting and storing information to better tailor our user experience, as per the Ngaro Terms of Service and Privacy Policy.

FAQ.

The questions people ask before they trust something with their documents. Anything missing, just ask us directly.

Is my data really never sent anywhere?
Your documents, your knowledge base and every conversation stay on your machine, on every tier. We never receive them, so we have nothing to leak, sell or be compelled to hand over. One honest caveat on the hosted tiers: when you ask something, your question and the passages your app pulls from your own documents are sent to be answered by a model we run, and they’re processed rather than stored. If you’d rather nothing left at all, self-hosted sends nothing during a conversation.
Do I need to know anything technical?
No. There’s no command line and no config file. You answer questions in plain English, upload files the way you’d attach them to an email, and install an app that works like any other app. It handles fetching the model for you. If you can install a program, you can do this.
Will it run on my computer?
Almost certainly. We check your hardware during setup and recommend a model that suits it. Machines with 8 to 16GB of RAM run our smaller recommendation comfortably; 16 to 32GB opens up the larger models for bigger document collections. If your computer genuinely can’t handle it, we’ll tell you honestly, and a hosted tier runs a bigger model for you while your documents stay where they are.
Can I change my AI after it’s built?
Yes. Add or remove documents, rename sources, change its name, tone or rules, and flip web search on or off, all from the settings panel inside the app. Teaching it something new happens on your device: reading and indexing a new document is done by the app on your own machine, and the file itself is never uploaded to us.
Do you train on my documents?
No, and we couldn’t even if we wanted to. We don’t retrain or fine-tune models at all. Your documents are stored in a searchable index and consulted at the moment you ask a question. That index never leaves your disk, on any tier, so there’s nothing for us to train on.