Introduction

The problem

An AI with access to your database writes SQL confidently and gets it wrong quietly. Not because it cannot write SQL, but because it does not know what “revenue” means in your company: whether it includes tax, whether refunds are deducted, whether it is counted on the order date or the payment date. That is not in the schema. It is in three people’s heads and a Slack thread from two years ago.

What qontexta is

A context layer: the place where those definitions are written once, reviewed the way code is reviewed, and served to everyone who asks — agents, BI, applications — with the same meaning.

Three decisions explain the whole product:

It lives in your Semantic Repository. The model is YAML files laid out as a standard Cube project in a GitHub repository that is yours. qontexta reads and writes them; it does not store them. If you ever leave, your model stays with you and keeps working: it is not a proprietary format.

The AI builds, you decide. The assistant reads your Data Platform, proposes cubes, measures and semantic views, and explains why. Nothing is written until you accept, and what you accept is written as a commit under your name. It is included — you do not bring an AI account.

Every definition has two halves. The technical one — where the data comes from, what SQL computes it — and the business one — what it means, what it excludes, who answers for it. Both live in the same file and each has its own sign-off. A metric is not ready until both are signed.

How it fits together

your GitHub repo  ←→  qontexta  ←→  your Data Platform (read-only)
   (model/*.yml)          │
                          ▼
             MCP · semantic SQL · REST · GraphQL

Concepts

Concept	What it is
Repository	Where the model lives. One per account, yours.
Connection	Read-only access to your data warehouse.
Cube	The technical entity: a table or query, with its joins, measures and dimensions.
Semantic View	The curated layer: what people and AI query (a Cube view).
Environment	A branch of the Semantic Repository, with its own model and its own outputs.

Next: the Quickstart.

Product concepts Quickstart