Private beta · invite only

Research that
compounds.

Most research dies at the readout. Umwelt turns interviews and documents into a structured body of evidence that keeps answering questions — months after the study shipped, and to people who were never in the room.

We onboard teams personally. No public signup yet.

Ask/OnboardingAAcme Research

What keeps coming up about onboarding across our last three studies?

Answer

Three studies, 23 sessions. Two themes recur, and they disagree with each other.

theme

Setup is re-entered per device — nine participants restarted onboarding on a second screen41117

problem

Errors fail silently — invalid input clears the field with no message719

Dissonance

The March study said onboarding length was the blocker2. The July study says length is fine and re-entry is the blocker14. Both are still live.

Ask a follow-up…⌘K
The problem

A finished study is a dead end

Every team has the same archive: a folder of recordings, a deck per study, and one person who remembers what's in them. The evidence exists. Nothing can reach it.

The synthesis is the bottleneck

Reading transcripts and pulling quotes is the slowest, least strategic part of the job — and it is the part that has to happen before anything useful can be said.

Findings don't connect

Study four contradicts study one and nobody notices, because the two live in different decks written by different people six months apart.

The answer is a person

“Did we research this?” routes to a researcher, who stops what they're doing to go and look. Institutional memory walks out with them.

What Umwelt builds

Five layers, and every one shows its work

Umwelt doesn't produce a summary. It produces a structure — each layer built from the one below it, each claim traceable down to the sentence a participant said and the second they said it.

atomic
Atoms

Individual observations, lifted from a transcript with the speaker, the timestamp and the surrounding context attached.

insight
Insights

What the atoms add up to within one study — proposed by the model, then accepted, edited or rejected by a researcher.

theme
Themes

Patterns that hold across studies. Each theme carries its participant count, so you can see how thin or how solid it is.

problem
Problems

The themes that represent something broken, separated from the ones that are simply true.

recommendation
Recommendations

What to do about it — still attached to the evidence chain, so anyone can audit the reasoning.

Not a summarizer

Anything that reads your transcripts and hands back prose is asking you to trust it. Umwelt is built the other way around: nothing is stated that can't be opened, and opening it lands you on the quote, the participant and the moment in the recording.

  • Every claim carries its sources — click through to the raw quote
  • Model output enters as a proposal, not as a finding
  • A researcher accepts, edits or rejects before anything is published
The product

Four places you'll actually spend time

Ingest

Drop in interview audio, PDFs, notes and transcripts. Audio is transcribed and diarized; documents are read as documents. Umwelt works out whether it is looking at a conversation or a single-author write-up and handles it accordingly.

mp3 · wav · m4a · ogg · flac · opus · webm · pdf · docx · rtf · txt · md · csv

Triage

A review queue of everything the analysis proposed, with its evidence alongside it. Accept, edit, reject. This is the step that makes the output yours, and it's the step most tools skip.

Ask

Put a question to everything you have access to, in plain language. Answers come back with citations you can open, and follow-ups keep the thread — so a PM can dig without booking anyone's afternoon.

Reports

Assemble findings into a readout and export it as HTML, Markdown, PDF or a slide deck. Counts in the export come from the same columns the app reads — a report never prints a number nothing backs.

Across the corpus

The part a folder of decks can never do

Dissonance detection

When a new study contradicts something already in the corpus, Umwelt raises it instead of quietly stacking both claims on the shelf. Contradiction is a finding, not an error.

Inflection signals & freshness

Themes carry an age. When one starts to move — new evidence, a shift in strength — it surfaces on your home screen rather than waiting for someone to go looking.

Who it's for

Same engine, different room

Individual researchers

Stop being the index of your own archive

A corpus you can question is worth more than a drive you can search. Umwelt gives a solo researcher the synthesis layer a whole team would otherwise have to build by hand.

Research teams

Everyone reads the evidence, not the deck

Shared corpus, shared taxonomy, and a review step that keeps standards intact as more people contribute. PMs self-serve; researchers stop being a lookup service.

Your data

Never used to train a model. Ever.

Research data is participant data. Umwelt runs on enterprise AI APIs under zero-retention terms, isolates every workspace at the database layer, and encrypts everything at rest and in transit.

Your best insight shouldn't expire with the readout

Umwelt is in private beta. We onboard teams one at a time so we can watch how it holds up against real corpora.