Questions we are asked.

What does Neural Logic actually deliver?

An AI colleague that reads the archives your company already holds and, for any case you ask about, lays out the complete file in date order, quotes the passages that decide it word for word with a link to the original, and states what it looked for and did not find.

Alongside it you get the archive itself: your files converted into a checked, addressable form that works with any language model, and a register of what each document contains.

Does our data leave Europe?

No. The colleague that reads documents runs on a European model, Mistral, processed in Europe under a European contract, and the work happens under a data-processing agreement with Axon Trade ApS, a Danish company.

The supervisor that improves the colleague overnight never sees a document or a name, so it is chosen for thinking rather than location: today that is Claude, from Anthropic, running outside Europe. We say so plainly rather than writing “the strongest model available”, because a boundary nobody can see is not a boundary anyone can check. It reads only the shape of the work — what was asked, how many documents, which word was missing — and never your material.

Do we have to migrate our documents or change systems?

No. A daily copy of the archives you already keep is enough: a data room, a mail archive, an accounting export, a shared drive. Nothing has to be moved into a new system, and your people keep working exactly where they work now.

What kinds of archives and files can it read?

Contracts, correspondence, invoices, protocols, reports and scans, across several archives at once. Scanned paper is read as text, so documents that were never searchable become part of the file like any other.

What happens when the answer is not in the documents?

You are told. “Not found” is a first-class answer here, and it comes from the register built while every document was read, not from a model's memory. Whether a clause, a signature or a protocol is missing is a fact about your archive, and it is answered as one.

For most of the work our clients care about, knowing that something is missing is worth as much as finding it.

How do you stop it from inventing quotations?

Every quotation is compared word for word with its source document before it reaches the reader. If it does not match, it is held back and the reader is told. Nothing reaches you on the strength of sounding right.

How is this different from asking ChatGPT or Copilot about our files?

A general assistant answers from what it retrieved and what it recalls, and it cannot tell you what it failed to find. This is the difference that matters in a file that carries money or liability.

A document colleague reads every document in scope, keeps a register of what is in them, quotes only what it has verified against the source, and names the gaps. It also keeps the boundary: your documents stay inside a European contract.

Who do we sign with, and who is liable?

Axon Trade ApS, a Danish private limited company in Copenhagen, register number CVR 45 92 07 63. Neural Logic is a brand of that company, not a separate legal entity. Contracts, invoices and the data-processing agreement all name Axon Trade ApS, and the company is signed for by its director.

What does a first engagement look like?

Small and measured. A short first contract on a real case of yours rather than a demonstration, with acceptance judged on what it produces; then two further short phases. You can stop after any phase and keep everything built so far, including the converted archive, the register and the glossary.

The point of starting this way is that you decide on evidence from your own files, not on slides. And the first phase is worth having even if you decide to stop there, for the reason in the next answer.

If we stop after the first phase, what have we actually got?

The thing every company will need anyway: your files converted into a checked, machine-readable archive, with a register of what each document contains and a glossary of your own vocabulary. It is not built for our software. It works with any language model, ours or whichever one you choose in three years.

Almost every company will have to do this work at some point. Documents that only exist as scans, PDFs and mail threads cannot be used by any AI worth trusting, and converting them is the slow part — not the clever part that comes after. Doing it on a small first contract, on real cases, with the result yours to keep, is a cheap way to have it behind you.

That is why we treat the first phase as a commitment rather than a trial: whatever you decide about us afterwards, you are further along than you were.

What does it cost?

It is priced per engagement, because the work depends on how much you hold and in what state. The first phase is deliberately small, so the first decision is a small one. Write to us with a rough idea of the archives involved and you get a straight answer.

Which languages does the colleague work in?

Danish, German and English. The company is in Copenhagen and its founder is German, so the written work happens in all three without translation in between.

Who is behind Neural Logic?

Thomas Möller, founder and sole owner of Axon Trade ApS in Copenhagen, which also operates the hockey shop Hockey24. The company facts, including the register extract, are on the Axon Trade page.