A small bridge on your network. Your database isn’t exposed.
The gateway lives next to your database and is the only thing that talks to it. It opens the connection outwards itself, over HTTPS, just as a browser would. There’s no inbound port to open.
Install the gateway
A container on a server in your network. You register it with a single-use code and a wizard that checks the connection.
Cuarzal studies your database
It reads the catalogue and aggregate statistics, without reading values, and flags the columns holding personal data. See how it does it.
Review and publish
You go through the business names and metrics it proposes. Before publishing, you see what changes.
Your team asks
Analysts and readers, each with their own permissions. Each person sees only what they’re entitled to.
Before answering, Cuarzal learns to read your database.
It does this in five layers. The first four are counts the gateway makes inside your network, with no AI. The fifth is proposed by the model without seeing a single row, and isn’t used until you approve it.
First, the structure.
The gateway reads your database’s catalogue: which tables there are, which columns they have, what type they are and what their keys are. It’s exactly what your IT person would see on opening it.
Then, how they relate.
Declared relations come from the catalogue. The ones nobody declared are inferred: if nearly all the values in one column exist in another, the relation is proposed along with its match percentage. That percentage is all that leaves your database.
Counts, not data.
How many rows each table has and, for each column, how many distinct values and how many empty ones. These are numbers about your data, calculated inside your database. No value leaves it.
Personal data, set aside.
It’s detected from the column name and from a sample of 1,000 rows checked inside your database: all that comes back is how many match. Emails, phone numbers, national IDs, IBANs, names, addresses and dates of birth stay blocked unless an administrator opens them.
And finally, the meaning.
From those four layers, the model proposes a business name for each table, the role of each column, and the metrics. Whatever it isn’t sure about, it asks you.
- PKFKown key · another table’s key
- declared relation
- inferred relation, with its %
- personal data, blocked
- fact table: what gets measured
- Sample database, made up · 15 tablesSample database, made up. Here, 4 of its 15 tables
You review
Table by table, you accept, edit or reject each proposal, and answer its questions.
You publish
Before publishing, you see what’s added and what changes. Only what’s published is used to answer.
If your database changes
Cuarzal detects it and sends only what’s affected back for review.
What’s needed from your IT person.
Three things. None of them means touching your network’s inbound configuration.
- In your database
A read-only user
It’s created by whoever administers your database, with the permissions they decide and a time limit per query. If there are columns Cuarzal mustn’t see, such as a national ID or a salary, simply don’t grant them: for the gateway, they don’t exist.
- On your network
A machine with Docker
One that can reach the database and get out over HTTPS, port 443. That’s where the gateway lives. It needs no inbound port, because it’s the one that opens the connection.
- From us
A registration code
It’s single-use and it expires. With it, a wizard in the browser tests the connection and shows you that user’s permissions report before anything is connected.
We’re looking for the first companies to try it.
If you have a database holding business information and want your team to ask it questions without waiting for anyone, tell us about your case. We’ll guide you through the installation.
Open in your email We reply in Spanish and English.