The researcher's workbench

From source to sense.

Interview transcription, source capture, summaries and qualitative coding – in one workspace. Slovene-first, processed in the EU, with no database.

Try the workbench Talk to us

The demo is open – no login, no sign-up.

The problem

Qualitative research leaks between tools

Your material crosses the world

Recordings of sensitive interviews land in US clouds that ethics boards and GDPR reviews struggle to approve.

Slovene is an afterthought

The big tools are built for English. Smaller languages get the leftovers – and the transcript shows it.

A different tool per step

Transcribe here, code there, summarise elsewhere. Copying in between, and the context is lost.

The workbench

Five tools, one flow

Pick the tools you need – one tool's output is the next one's input, in a single click.

Transcription

Meetings and interviews, uploaded or live. Timestamps, speaker separation, Word export.

Qualitative coding

Codes and sub-codes, analytic memos, coded extracts to Word or the codebook to CSV.

Source capture

Readable text from a public page – an article, a report – ready to analyse.

Summaries

Summarise any text or transcript, with action items.

Paraphrasing

Clearer, shorter, more academic – in Slovene and English.

For institutions

Your brand, your domain, your rules

Scriptorium is white-label: your institute gets its own build of the workbench – your logo, your colours, your subdomain.

Your visual identity

Logo, colours and name on your subdomain. Your researchers see your tool, not ours.

Choose your modules

An institute buys transcription + coding; a language department might take only the writing tools.

Sign in with SSO

Connects to your existing identity provider (Keycloak / OIDC), so you control access.

Processed in the EU

The app runs in Frankfurt. Where each model runs is disclosed publicly and precisely.

Integrity

Built for publication, not for the demo

AI proposes, the researcher decides

Auto-coding offers suggestions; nothing is applied until a human accepts it.

Honest model provenance

Every summary carries the model that actually produced it – not the one we wish had.

Open weights, properly credited

Slovene and open models are cited with the acknowledgment their authors require (CJVT, ARIS, NextGenerationEU).

No quiet storage

The transcript lives in your session. Audio is only in transit. Codes stay on your device.

FAQ

What you'll ask first

Where is data processed?+

The app runs on EU servers (Frankfurt). Some language models currently run at US providers – this is disclosed precisely on the »About the models« page, including which model does what. The goal is a Slovene model hosted in the EU.

Is anything stored?+

Not server-side. The transcript lives in your browser session; audio passes through transit and is deleted immediately; codes are saved only on your device. Institutions wanting persistent projects get that as an explicit opt-in, never the default.

What about Slovene?+

Slovene is the first language, not a translation. The workbench is designed around Slovene open models (the GaMS family, CJVT), with other open models as the fallback when those are not running.

Can we export to NVivo or MAXQDA?+

Coded extracts export to Word, the codebook to CSV. The first-phase goal is a clean handoff into your existing tools, not replacing them.

What does it cost?+

It depends on the modules and the size of the institution. Write to us and we'll give you a real number, not a guess.