nlqdb

AI can’t replace you if you become AI.

We’re building a marketplace where your professional knowledge becomes a structured database — typed rows agents query in plain English and, when the marketplace opens, pay you to use. Not a chatbot trained to sound like you: a knowledge base that stays yours.

In build, in the open. Everything below says whether it is shipped today or still being built — nothing on this page claims more.

How you become AI

You talk. It becomes a database. Agents query it.

1 · You talk in build

No schema, no SQL, no forms shaped like a database. You answer an interview in plain language; each answer becomes typed rows.

“A student says “yo sabo”. What is happening, and what do you do next?”

facts
├─ kind      error_pattern
├─ content   regularises irregular first-person
│            (sabo → sé); high-frequency verbs
├─ tags      {beginner, verbs, first-month}
└─ source    interview

Design fixture — the interview product is in build; this shows the designed shape, not a live product.

2 · It becomes a database engine live

A real Postgres database of what you know — typed rows you can see, correct, and delete. The engine that runs it is live today: it powers analytical memory for AI agents now.

“Which mistakes do beginners repeat most in their first month?”

SELECT content AS mistake,
       COUNT(DISTINCT end_user_id) AS students
  FROM facts
 WHERE kind = 'error_pattern'
   AND 'first-month' = ANY (tags)
 GROUP BY content
 ORDER BY students DESC;
mistakestudents
ser vs estar in past descriptions21
regularises irregular first-person (sabo → sé)17
gender agreement on -ma nouns (el problema)12
por vs para for deadlines9

Rendered from a fixture in the shipped table shape (agent_memory_v1 on Postgres); the English→SQL compile is the live engine. Row values are illustrative.

3 · Agents query it — on your terms in build

When the marketplace opens, another agent reaches your knowledge only through a grant you create and can revoke — and buyers pay to query it. Until that ships, no one’s agent but your own can touch your rows.

The trust deal

Not ours to read.

The deal the marketplace opens with: nlqdb is not allowed to use or read your knowledge — it is only yours. That is a contractual prohibition, written down before the first listing goes live, standing on the floor below — with nothing claimed beyond what is shipped.

The contract comes first.
Before the first listing goes live, our terms add a plain prohibition: nlqdb may not read, use, mine, train on, or resell your knowledge for any purpose other than running the queries you and the buyers you authorise make. The marketplace does not open before those terms do.
What the engine necessarily does.
To answer a question, the engine reads the relevant rows on the server — that is the service you turned on. What the contract rules out is every other use.
Planning is schema-only.
When a question becomes a query, what reaches the language model is your data’s structure — table and column names, types, the descriptions you author — not your rows.
Narration is disclosed, and skippable.
When an answer is narrated back in plain language, the rows that answer returned do pass through the model. That is disclosed in our privacy policy today, and API callers skip narration by requesting JSON-only responses.
Isolation by default.
Your rows sit behind Postgres row-level security, isolated per account.
Our models don’t train on your data.
Already our written privacy commitment for every database on the platform: we don’t use your data to train our own models.
Delete means delete.
You can delete a database at any time; retention timelines are in the privacy policy.
Run it yourself.
The engine is source-available under FSL-1.1 — you can run it on your own infrastructure for any non-competing use.

The written record: privacy policy · terms · manifesto.

The pilot

First profession: the language tutor.

We’re building the whole journey on one profession before opening it wider: a language tutor’s error taxonomies, mistake patterns, and progress signals — knowledge that is analytical by nature, which is exactly what a structured database is for. The founder is user #1: the first knowledge database is authored and reviewed in-house, before any stranger’s livelihood depends on the platform.

What you can do today

Try the engine →

Create a database from one sentence and question it in plain English — anonymous, no sign-in, gone in 72 hours.

Building agents? →

The same rows power analytical memory for AI agents right now — memory your agents can GROUP BY.

When the authoring interview opens, it starts on this page — no email to leave in the meantime.