How it works
From a conversation to company memory.
Four steps, each with a human in the loop. No black box, no scraping: a deliberate pipeline from expert to answer.
Interview
An AI interviewer asks the questions only your experts can answer: voice-first, 30 to 60 minutes. It captures the reasoning, not just the result.
Review
Every transcript is read and approved by your team before anything is stored. Nothing enters the company memory without a human saying yes.
Forge
Approved answers become typed Knowledge Cards. Classified, sourced, and tagged with who is allowed to see them.
Retrieve
Your team asks in plain language. Answers come back with citations, or an honest “no source” when the knowledge simply isn’t there yet.
Why it’s different
Built to be trusted, not just impressive.
Most tools retrieve the documents you already have. The hard part (and the point of Legacy AI) is the knowledge that lives only in people’s heads, governed so you can actually rely on it.
Cited, or nothing.
Every answer carries its source. When there is no grounded source, Legacy AI refuses instead of inventing one. Honesty is the default, not a setting.
Permission-aware by construction.
Who can see what is enforced in the database itself (Postgres row-level security), not bolted on in application code that can be bypassed.
It captures what was never written down.
Interviews surface tacit expertise: the judgment and reasoning behind the work. Search tools only index what already exists; we raise what doesn’t.
EU-native.
Hosted in Germany, GDPR-first. Export and deletion per person, an append-only audit trail, and a data-processing agreement on request.
Why now
A generation of expertise is retiring at once.
The most experienced people in industry are leaving the workforce faster than they can be replaced. What they carry is mostly undocumented, and once they’re gone, it’s gone.
“Roughly 70% of what an organisation knows is tacit, held in people, not documents.”
Documents capture what people did.
Never why they did it. The rationale, the edge cases, the “don’t ever do that”: that’s the part that walks out the door.
A wiki is written by whoever has time.
Which is no one. Exit interviews are an HR ritual, not a knowledge transfer. The result is the same either way: loss.
AI is only as good as the corpus you feed it.
Point a model at your PDFs and it returns fragments without context. Feed it captured expertise and it can actually reason.
The honest part
The most valuable answer is sometimes “I don’t know.”
A model that always answers will always sound confident, even when it’s wrong. Legacy AI would rather tell you the truth.
FAQ
The questions we get asked.
Short answers, in plain language. If yours isn’t here, email us and we’ll answer it properly.
What does Legacy AI actually do?
It interviews the people who hold your critical know-how, before they leave. A voice-first interview draws out the reasoning behind the work, a human reviews and approves what was captured, and the result becomes typed, cited knowledge your team can ask questions of. The category is knowledge retention; the established name for the manual version of it is expert debriefing.
What happens to an expert’s knowledge when they retire or leave?
Today, most of it goes with them. Handover documents and the usual offboarding checklist capture the what, rarely the why. Legacy AI treats a retirement or a resignation as a knowledge event: the judgment, the decisions and the workarounds are captured while the person is still there to explain them, and stay available to whoever picks up the work. It applies to succession planning and shift handover just as much as to a departure.
Does this replace our wiki, SharePoint or knowledge management system?
No. Those hold what somebody already wrote down, and Legacy AI does not compete with them. It is for the other half: the expertise that was never written down at all. Search tools can only index what exists; the point here is to raise what doesn’t.
How do you stop it from inventing answers?
Every answer carries its source. When there is no grounded source, Legacy AI says so and refuses, instead of producing something that merely sounds right. That refusal is the default behaviour, not a setting you switch on.
Where is our data hosted?
In Germany, on EU infrastructure, GDPR-first. Export and deletion per person, an append-only audit trail, and a data-processing agreement on request.
Who can see what?
Permissions are enforced in the database itself, through Postgres row-level security, rather than in application code that can be bypassed. An answer is assembled only from the sources the person asking is cleared to see.
Start preserving
The expertise stays. The person moves on.
Bring us the knowledge you can’t afford to lose. We’ll show you how a single interview becomes memory your whole team can trust.