Data foundations for AI
The meaning of your data comes from your business.
The foundation AI needs is a precise understanding of what your data means to the people who run the business. We capture it from the people who make the decisions, in their own words, and keep it as rules anyone can read and check.
The business here is a sample, so you can try it without handing over any data.
Starmaker
One model holds the logic of your business, and Starmaker builds everything else from it.
That logic is what each thing in your business is, how the things relate to one another, and what must always be true about them. Today most of it lives in people's heads and in hand-written code spread across pipelines, where nobody can see it whole. Starmaker writes it down with the people who know it, the people who make the decisions and the person who knows your database, and uses AI to take that conversation to a loaded, checked star schema in a single working session.
1 · Your logic
The people who make the decisions describe what they measure and how the business works, in their own terms. AI drafts the model as they talk, and they correct it until it is right.
2 · Your data
Starmaker reads your database and proposes where each part of the model lives, whether stored or calculated, and the person who knows the database decides. Every proposal is measured against your rows before anyone accepts it.
3 · Your star
Starmaker designs the star schema from the model, generates the transformation process that loads it, and checks the loaded star against your source, total by total.
4 · Your numbers
Pick what you want to see, by customer, by month, by whatever the question is. Every number adds up the way your business said it does, and any number opens to the transactions behind it.
Each product belongs to exactly one subcategory.
This is business logic, and the model is made of rules like it. Anyone in the business can read one and say whether it is true, and the star, its load, every report and every AI answer follow it.
Because everything is built from the model, it holds up as the business changes. A change to the logic is made once, in the model, and the star and its load are rebuilt from it. Every load is checked against your source, every version of the model is kept and can be restored, and anyone with access can see how each table is made. Starmaker reads only what the model needs from the database where your data already lives, so the size of your source does not set the size of the work. And because the model is written in GRAM, Claude and any tool connected through MCP query the star by the same logic.
For the data team
A complete platform around the model.
GramSpec connects to the databases you already run and keeps the model where your team can manage, secure and audit it.
- Connects to SQL Server, PostgreSQL, MySQL, MariaDB, Snowflake and Databricks, or a CSV upload.
- Imports the structure of a live database or a Power BI template.
- Keeps every version of the model, with a diff and a restore.
- Names a steward for each table.
- Restricts rows for each person who can see a board.
- Runs read-only, with encrypted credentials, two-factor sign-in and a log of every query.
- Answers and investigates on the same rules, in Vision and Reason.
- Writes the model in GRAM, from the first conversation to the loaded star, and hands it to Claude, Cursor and your own apps through the connector and an API.
Read the specification API reference The research paper (PDF)
Build the foundation from what your business knows.
Bring the people who make the decisions and the person who knows your database.