Faster studies, and better evidence behind them.

The repository solves storage. It does not solve the two hard parts.

A research team’s throughput is not limited by how fast they interview. It is limited by the hours on either side: turning a vague idea into assumptions that can be tested, hypotheses that can fail and a guide somebody can run on Thursday, and then turning forty highlights into something a roadmap conversation can use.

That is where discovery quietly stops happening. The study gets postponed because nobody has an hour to write the guide, or it runs and the findings sit in a deck nobody opens again.

It makes researchers faster, and more people able to help

This replaces no one. A researcher still decides what is worth asking, still runs the conversation, still judges what the evidence means. What changes is how much of a week that takes, so one researcher covers more ground and the backlog of studies waiting on a guide stops growing.

The same thing opens discovery to the people around them. A product manager or designer can run a study inside the structure a researcher would have given them, with the hypotheses written so they can fail and the analysis plan already in the guide. The researcher reviews and coaches, and stops being the only route to a customer conversation. That is what research democratization actually requires: not fewer researchers, but a floor under everyone else.

Every claim stays two clicks from its evidence

Plan. Enter the assumptions and hypotheses in whatever shape they are in. They get categorized by desirability, viability, feasibility and usability, each hypothesis sharpened into a form that can be proven wrong, a method recommended, and a complete guide written with screener, interview script and analysis plan.

Tag. Work through an interview and mark the passages that matter. Tag suggestions arrive per highlight with a tick box, each card showing what the repository already holds for that passage: new, adds tags, already complete, overlaps an existing highlight.

Insights. A cluster of quotes becomes a headline, a description, evidence with confidence levels, counter-evidence, and the questions still open.

Opportunities. Insights map into an opportunity tree, and the assumptions for the next study load straight out of it, which closes the loop that usually has to be reopened by hand.

A study holds all four. Every insight keeps a line back to the quotes it came from and the plan it answered, so the thing a stakeholder challenges is always two clicks from its evidence.

Nothing arrives in your repository twice

The part that matters most is the least visible. A quote that matches a highlight somebody already made, by hand or otherwise, adds its tags to that highlight. No second copy of the sentence arrives. A proposed tag that already exists is reused. The glossary is checked against what the repository actually holds, including its trash, and a refused tag comes back with the reason and the fixes to try.

This is the work nobody demos and everybody needs, because a tool that pollutes the repository gets switched off in a month.

Research that already exists somewhere else does not have to be retyped. A screenshot dropped onto a step is read, and the text lands in the fields.

Built by someone who had to solve this at work first

The method is continuous discovery as Teresa Torres describes it, and the reason it is built this way is that I have run the alternative. At FTI Group I built a research function from zero to eleven people and watched capable product teams stall on the same first step every time. At Conrad Electronic the function is five people serving seven product teams, which only works when the teams can run discovery themselves.

A workspace is what remains when you cannot put a researcher in every team. The product exists because I needed it twice.

Nothing follows the user

No analytics, no cross-site tracking and no consent banner. Turnstile guards the API without tracking anyone across sites. The Anthropic API key never reaches the browser, and the client chooses neither the model nor the prompt. Sign in with a Google account or an email link, whichever your company permits.

For teams in the EU there is a data processing agreement, alongside a German imprint, privacy notice and terms, because a research tool that cannot answer the procurement questionnaire never gets adopted.

It ships weekly, and the changelog is public

The workspace is in active development and the changelog is open. Recent weeks have gone almost entirely into the repository integration, which is the kind of detail that decides whether a tool survives contact with a real workspace.

What it says about how I lead

I still build. A product leader who ships a working AI product end to end, with the security, privacy and integration details right, argues from a different place in a roadmap discussion than one who can only commission it.