Memory

A memory with receipts.

A language model that has not read your documents will invent them. So we built the memory first: courts, statutes, registers and public records, cleaned, linked and indexed, so that every answer can point at the page it came from.

-vectors in memory
-court decisions indexed
-decision-to-statute links
-embedded in the new index

From a document to a fact

CRAWLOfficial sources, politely, on a schedule.
CLEANLayout noise removed, text normalised, duplicates folded.
LINKEach decision is tied to the statutes it applies.
EMBEDMeaning becomes a vector, stored on our own hardware.
CITEEvery answer carries the identifiers it used.

Why it matters

01

Law and courts

Court decisions and statutes in one graph. A question about a rule returns the rule, the decisions that applied it, and where they disagree.

02

Public records

Registers, procurement and civic data resolved into entities, so that "the same company" is the same company everywhere.

03

Two index generations

The memory is being re-embedded with a newer model while the old one keeps serving. Nothing goes dark during the upgrade.

04

Sovereign by default

Documents and vectors live on infrastructure we operate in Europe. Retrieval needs no third party.

Numbers on this page are read from the live system every thirty seconds.

Free access

See the live system.

Register once, free, and the architecture and live demo open for 30 days. Payment is only for the reference code.

Register — free →
Behind the door

Read how it is built.

Six reference modules in textbook form: access, telemetry, payment, self-improvement, grounded retrieval, consent-first measurement.

Open the door — EUR 100 incl. VAT →
RiNET · Engineered in the Netherlands · Operated on sovereign European infrastructure · Privacy