Research needs more than plausible answers.
Language models can be remarkably good at explaining, summarising, and writing. But research has a different requirement: claims need to be verifiable.
Why we build it this way
That is why we are not building Acadex as a chatbot that simply sounds convincing when it talks about research.
Acadex combines language models with bibliographic data on publications, researchers, and citations — so answers can lead back to something you can inspect.
We do not want to replace what researchers need to judge for themselves.
Acadex can help uncover connections, work through large bodies of literature, compare research profiles, and make bibliographic data easier to explore. It should not decide what good research is.
A highly cited paper is not automatically a better paper. A frequently cited researcher is not automatically the better researcher. And an answer from a language model does not become correct simply because it is confidently written.
Acadex should give you more evidence to work with. The judgement remains yours.
Traceability is not an extra feature.
When Acadex derives an answer from bibliographic data, the underlying publications and records should remain visible. Not because that makes every answer automatically correct, but because a research tool should let you check for yourself.
If a data source does not cover something, Acadex cannot make that information appear. If a model gets something wrong, the underlying records should still give you a way to inspect where the answer came from.
We would rather expose those limits than hide them behind a polished response.
The model is only part of the system.
Acadex is not one language model with a research-themed interface around it. It is an agentic research system that brings together language models, bibliographic data, search, calculations, document processing, tools, and persistent context.
The underlying model can change depending on the task or configuration. The surrounding system is what determines what data is used, which tools are available, how results are calculated, and what can be traced back afterward. That distinction matters to us.
Acadex is not finished. Right now, that is intentional.
Acadex is still in Early Access. Some fundamentals are already clear. Others should be shaped by how the product is actually used.
We do not want to build every possible feature simply because it can be built. We would rather understand where research work actually becomes difficult, repetitive, or hard to verify — and work from there. That also means saying no to things that would make the product busier without making the work better.
We do not want advertising, engagement mechanics designed to keep people inside the product, or the use of private research work to train our own models. What we build should come from real research workflows — not from the longest possible feature list.
We measure whether the product works. Not what people are researching.
Building without any usage data would mean building in the dark. So Acadex collects a small number of product and business metrics on its own infrastructure. That can include registrations, activation, plan changes, cancellations, and aggregated page views.
We do not use conversation content, uploaded documents, saved memories, research topics, or bibliographic search queries for product analytics. We also do not use external advertising or behavioural analytics platforms for this purpose.
We want to understand whether Acadex is useful — not build a profile of someone’s research interests.
Acadex is built in Osnabrück.
We develop the product in direct contact with the people who use it. If something is missing, an assumption is wrong, or Acadex does not work well for your field, we want to know.
That feedback goes directly into the decisions we make about the product. Especially during Early Access, that is some of the most useful information we can get.
We would rather say “not yet” than pretend.
Some things Acadex does not currently have. That includes parts of the institutional stack such as individual-user MFA, institutional SSO, a general SLA, and external security certifications.
We document those gaps instead of treating them as marketing details to be hidden. If a missing capability matters for a real deployment or research workflow, that is useful information for us. It may change what we prioritise next.
If you want to know why something works the way it does, ask us.
Some things can be explained on a website. Others are easier to discuss directly.
Whether it is about the product, an institutional deployment, coverage of a research field, a privacy or security question, or simply a critical question about Acadex — we are happy to hear it.
Further reading
Whatever you need, there is a person at the other end
We are a small team in Osnabrück. Pick the route that fits and you will reach someone who can actually answer.
Universities & institutions
Seats for a group or department, payment by invoice, a data processing agreement, or a walkthrough with your team.
Talk to usHelp with your account
Something is not working, or you want to know whether Acadex can do the thing you have in mind.
Get supportAnything else
Press, partnerships, or a note about how well your field is covered. That last one is genuinely useful to us.
Send a message