The internet has been read. The archives have not.

Company

TRIVDA licenses those archives from the organisations that own them, puts them in a form an AI system can use, and enforces the owner’s terms on every request that reaches them. That is the whole company.

Entity: TRIVDA AB
Based: Stockholm, Sweden
Hiring in: Stockholm and London

A single aisle between tall steel library stacks, shelf labels overhead, running back to a far wall of books. — Photograph Zoshua Colah

Why this company exists.

01
A model gets better by being fed

The recipe has not changed since OpenAI published Scaling Laws for Neural Language Models in January 2020: a model improves when parameters, computing power and training data grow together. The first two can be bought. The third had to come from somewhere, and for the first generation of models it came from the open web.
02
The open web has already been read

There is only one of it, and the models have been through it. The knowledge that would make the next generation better was never on it to begin with. It sits in research archives, technical manuals, case files, back catalogues and operating records, inside the organisations that wrote them.
03
Those archives are shut, and reasonably so

They are written in formats no model can read, and they belong to someone. Opening them is therefore not a scraping problem. It takes a licence the owner agrees to, a record of exactly what that licence permits, and a check on every request that reaches the data.
04
That check is the company

TRIVDA licenses an archive at its source, makes it machine-readable, enforces the owner’s terms on every query, and pays the owner each time their data answers one. We are not a broker, not a scraper, and not another model company.

How it is enforced

Stockholm from the water: Riddarholmen church spire over the rooftops of Riddarholmen and Gamla stan, a white steamer moored along the quay. — Fig. 01 Based in Stockholm, Europe Riddarholmen and Gamla stan, from the water Photograph Johan Anblick

We work on the near shore of the water in that photograph.

Being a European company is not incidental. The rules here start from the assumption that data belongs to someone, and that using it means asking first. So does TRIVDA. Regulation is the shape of what we built, not an obstacle we built around.