NASA lost the ability to rebuild the Saturn V rocket. Not because the blueprints vanished. But because the engineers who knew how to read them retired. Your organization has the same problem. And it's happening faster than you think.
In the spring of 1961, President Kennedy stood before Congress and declared that America would put a man on the moon before the decade was out. The country had virtually no idea how to do it. What followed was the greatest engineering sprint in human history: the Apollo program, powered by the Saturn V: a 36-story tower of aluminum and ambition, five F-1 engines burning 15 tons of fuel per second, producing more power than 85 Hoover Dams combined. It was, and remains, the most powerful machine ever flown.
The Saturn V flew 13 times. It never lost a single payload. It put 12 human beings on the surface of another world. And then, in 1973, after the final Skylab mission, the production lines stopped. The tooling was scrapped. The engineers went home.
Four decades later, in 2013, a new generation of NASA engineers decided they wanted to understand the F-1 engine, not just read about it in textbooks, but actually grasp how it was built. They had every blueprint. Thousands of pages. Meticulously preserved. Every bolt, every weld, every tolerance annotated in ink that had barely faded.
They couldn't build it.
The plans showed what to build, but not how. The weld technique that needed "a bit more heat in winter" — a note the original welder never wrote down because everyone in the shop just knew. The alloy substitution the metallurgy team discovered after three catastrophic blowouts on a test stand in Huntsville, Alabama, in 1966 — a fix that lived only in the memory of the engineers who were there. The test procedure the senior propulsion lead always ran unofficially, because the official one missed a pressure spike that had nearly killed two technicians. None of it was documented.
The blueprints were perfect. They were also useless.
NASA — with the largest budget of any scientific organization on Earth, with access to the brightest minds on the planet, with the original hardware literally sitting in museums, had to reverse-engineer its own technology like archaeologists trying to reconstruct a lost language from pottery shards.
They succeeded, eventually. It took years. It cost millions. And it could have been avoided entirely if the knowledge hadn't walked out the door with the people who carried it in their heads.
This is the problem ThreadWeave exists to solve. It's an open-source tool that connects to the systems your organization already uses: email, chat, documents — and builds a searchable memory of decisions, answers and scattered knowledge that would otherwise be lost when people leave. No forms to fill in. No wiki to maintain. It runs on your own hardware, passively weaving the threads of institutional knowledge into something durable.
It's free. It's AGPL-licensed. And it's built to catch what your organization knows before it retires.
To understand why this matters, let me tell you about Lars.
Every firm has a Lars. Not necessarily named Lars. Maria, David, Priya, whoever. The person who's been there since before the merger. The one who knows why the Stockholm office uses a different engagement letter structure than Copenhagen, even though nobody can find a written policy about it. The one who was in the room in 2008 when the big client almost walked, and knows exactly which partner said exactly the wrong thing, so nobody makes that mistake again.
Lars is the institutional memory. He's the reason a junior consultant can walk into a meeting with a client they've never met and somehow know about the sensitivity around the Q3 numbers. He's the person the new hires are told to "go ask." He's the safety net.
Then Lars retires.
And someone opens a file from 2014 and stares at a spreadsheet that makes no sense, because the logic behind it was explained over coffee once, eight years ago, and nobody wrote it down. A client mentions a precedent from a project in 2011, and three senior people nod along, and not one of them actually remembers the details. But Lars would have. A competitive bid hinges on a methodology the firm developed in 2006, and the only person who could explain why it works hasn't worked there in a decade.
This isn't a filing problem. It's not a documentation problem. It's an existential risk that compounds silently, invisibly, until one day it isn't invisible anymore.
There is a demographic wave coming that makes this urgent in a way most organizations haven't reckoned with.
In the professional services sector — law, consulting, accounting, engineering, architecture — the numbers are stark. Over a third of partners and senior practitioners are baby boomers. They are not "thinking about" retirement. They are actively doing it. The American Bar Association has been sounding the alarm for nearly a decade. The accounting industry's own professional bodies call it a "talent cliff." Consulting firms are running internal succession-planning task forces that produce reports nobody reads.
Here's what the data doesn't capture: when a senior practitioner retires, they don't take a filing cabinet. They don't take a laptop or a set of document templates. They take the thing that made them senior in the first place: the pattern recognition built over 30 years of hard-won experience. The ability to look at a client situation and say "this reminds me of something from 1999", and be right. The instincts. The judgment. The stories.
And here's the part nobody talks about: the people replacing them are entering a different world. Junior practitioners today are more digitally fluent, more efficient, more comfortable with technology than any generation before them. But they also job-hop at rates that would have been unthinkable a generation ago. The average tenure of a junior associate at a large firm is now under three years. So even if you somehow convinced every senior partner to document everything they know before they leave, and you won't, because they're busy, and nobody got into this profession to write wikis, the people you're documenting it for might not be around long enough to absorb it.
The Thomson Reuters legal blog put it bluntly: "When experienced staff leave, they take with them their best practices. If best practices exist only in the employees' minds or scattered on sticky notes, the chain of knowledge is irretrievably broken."
The chain is breaking. Right now. Across thousands of firms, consultancies, and professional practices. And the dominant response — across the entire sector, is to do nothing and hope the problem solves itself.
You have systems. Everyone does. SharePoint sites that nobody can navigate. Document management systems where files go to die. Microsoft 365 Copilot, if you're ahead of the curve.
Here's the uncomfortable truth: Copilot can only find what someone typed into a document.
Copilot is an extraordinary piece of technology. It can search your emails, your Teams messages, your document library, and return answers in natural language. But it is searching the explicit record, the things people wrote down. It doesn't know that a precedent from 2012 applies perfectly to your current case unless someone authored a memo saying so. It doesn't know which clients are unhappy because of something that happened in a meeting three months ago unless someone typed meeting notes. It doesn't know the unwritten methodology your Oslo office uses that's subtly different from Stockholm's. The thing everyone just knows — because it was never written down.
Copilot is a brilliant librarian in a library missing half its books.
The tacit knowledge — the context, the connections, the stories, the instincts, the stuff people carry in their heads, is completely invisible to it. And in professional services, tacit knowledge is the product. It's what clients pay for. It's the difference between a firm that can bill premium rates and one that competes on price.
Microsoft knows this. They've been trying to crack knowledge management for two decades. SharePoint Areas. InfoPedia. Viva Topics — launched in 2022 with fanfare and quietly killed in 2025. Each attempt failed for the same fundamental reason: you cannot capture tacit knowledge by asking people to fill in forms. Nobody wakes up excited to document what they know. They wake up excited to do the work. The capture has to happen passively, in the background, as a natural byproduct of doing the work itself.
The knowledge management industry has been writing books about this problem for three decades.
Search "knowledge management" on the Kindle Store and you'll find 2,500 titles. The top results, sorted by relevance: a beginner's guide, an AI wrapper book, a personal productivity system, a handbook with 133 reviews, a US Army field manual, and a sponsored ad — because Amazon couldn't find a sixth KM title worth showing organically.
The field's bible is Kimiz Dalkir's Knowledge Management in Theory and Practice, published in 2005. 372 pages. Nine reviews on Amazon. Chapter 4 — "Knowledge Capture and Codification" — teaches you to interview experts, draw cognitive maps, and build taxonomies. Chapter 8 — "KM Tools" — covers blogs, wikis, and data mining. This was cutting-edge KM technology in 2005. The book is still the #2 result on Kindle today.
Three decades. 2,500 books. Zero infrastructure. The entire field has been selling methodology — here's how to think about knowledge — while the problem sits silently in your inbox, growing larger every time someone retires.
Let me offer a reframe — one that matters if you've ever tried to explain this to a partnership board or a leadership team:
You insure the building against fire. You carry professional indemnity insurance. You have cybersecurity insurance. You probably spend more on insurance premiums than you'd care to calculate.
You do not insure against your organization forgetting what it knows.
And yet the financial cost of that forgetting — in duplicated work, in missed opportunities, in client relationships that wither because the person who understood them left, in competitive bids lost because nobody remembers how the firm solved a similar problem in 2011 dwarfs every insurance premium you pay. By an order of magnitude.
This isn't a software purchase. It's an insurance policy. The premium is attention. The coverage is knowing that when Maria retires, Maria's knowledge stays — not in a dusty wiki nobody reads, but woven into the fabric of how the organization works, accessible to the people who need it, when they need it.
ThreadWeave connects to the tools you already use — Microsoft 365 or Google Workspace, your email, your documents, your meeting transcripts, and builds a knowledge graph passively, in the background. No one has to change how they work. No one has to fill in a form. The system listens, connects, and weaves the threads of institutional knowledge into something durable.
It runs on-premises, inside your walls. Your data never leaves. There is no cloud vendor who can change their pricing model and hold your organizational memory hostage.
And crucially, ThreadWeave doesn't compete with Copilot or Gemini. It extends them. By feeding ThreadWeave's knowledge graph into your existing AI assistant's search surface, it suddenly gains access to the tacit knowledge it was blind to. It can answer "what was the approach we used for the Oslo project in 2019?" — not because someone wrote a memo, but because ThreadWeave connected emails, meeting transcripts, and document edits into a coherent thread of institutional memory.
Every thread, woven into memory.
Most knowledge tools dump everything into one flat, searchable pile. That works for a team of five. It fails for an organization of five hundred, because knowledge isn't flat. It belongs to people, to teams, to projects, to moments in time.
ThreadWeave is built on MemPalace, an open-source memory system that models knowledge the way organizations actually work. Think of it as a building:
The Palace is the entire organization. Every piece of institutional knowledge lives here.
Wings are the departments and teams: tax law, M&A, the Oslo office, the platform engineering group. Each wing holds the knowledge that belongs to that part of the organization.
Rooms live inside wings. A room might be a specific client engagement, a regulatory domain, a product launch, a recurring project type. This is where the actual content lives — the emails, the meeting transcripts, the document edits, the decisions made in a Teams thread at 11 PM that turned out to be important.
Drawers hold the raw material — exactly what was said, verbatim, never altered. Closets hold compressed summaries, refined over time as more context accumulates.
This structure matters because it mirrors how people actually think about knowledge in an organization. When someone asks "what was the approach we used for the Oslo project in 2019?", they're not asking for a keyword match. They're asking for something that belongs to a specific wing, a specific room, a specific moment in time. The structure makes the answer findable.
A flat structure would keep knowledge locked in silos. The tax team's insights stay in the tax wing. The Oslo office's lessons stay in Oslo. This is exactly the problem we're trying to solve.
MemPalace adds a knowledge graph that independently tracks relationships — who created what, which project it belonged to, which team was involved, when it happened. These aren't tags someone has to apply. The system builds them automatically by watching how information flows through the organization.
Hallways connect knowledge within the same wing. Everything the M&A team has ever done is linked.
Tunnels connect knowledge across wings. The tax team solved a regulatory classification problem in 2018 that the M&A team is about to encounter in 2026. Without a tunnel, nobody would ever know.
Tunnels are ThreadWeave's answer to "what did Lars say?" Not because anyone explicitly documented the cross-department connection, but because the system traced the information as it moved through the organization and built the link automatically.
Not all knowledge is equally relevant forever. A regulatory interpretation from 2018 might be perfectly current. A technical decision from 2018 might be ancient history. MemPalace models this with Living Memory — connections that get used grow stronger, connections that go unused gradually fade. The knowledge graph is temporal, it knows when things were true.
When someone searches for knowledge, ThreadWeave weighs four factors: semantic match, organizational proximity, freshness, and author authority. The result isn't a list of documents. It's an answer that feels like institutional instinct.
| M365 Copilot | ThreadWeave |
|---|---|
| Searches what's explicitly written | Captures what's implicitly known |
| Finds documents matching your query | Builds connections between people, projects, and knowledge |
| Requires someone to have documented it | Works passively, in the background |
| Cloud-dependent | On-prem — your data never leaves |
| Proprietary, per-seat pricing | Open source, AGPL, free forever |
ThreadWeave wraps MemPalace — the highest-scoring open-source AI memory system — which provides the hybrid semantic+keyword search, the palace/wings/rooms architecture, the knowledge graph, and the temporal memory model. We're grateful to stand on that foundation.
ThreadWeave is AGPL-licensed. Not because we couldn't think of a business model, but because organizational memory is infrastructure. It shouldn't belong to a vendor. It shouldn't have a per-seat price that multiplies when your firm grows. It shouldn't vanish if the company behind it gets acquired or pivots to "AI-powered something-or-other."
AGPL is a legal commitment. Nobody can take this code proprietary. Not us, not a competitor, not a private equity firm that buys the repo. Your organization's knowledge belongs to your organization.
Free forever. For the global consultancy with 10,000 people. For the two-partner boutique. For the accounting firm in Bergen that's tired of asking "what did Lars say?"
If you lead an organization where expertise is the product: Take ten minutes. Walk down the hall. Ask the person who's been there the longest: "What do you know that nobody else knows?" If the answer makes you uncomfortable, good. That's the starting line.
If you're responsible for technology: ThreadWeave runs on your existing M365 or Google Workspace stack, on-premises. It extends Copilot instead of competing with it. Open source. Free.
If you're an engineer: The repository is open. Infrastructure-level work — the kind that matters decades after you ship it.
If you recognized the problem in your own field: Tell us what the concrete knowledge management challenge looks like in your world. Reach out. Open an issue. Help us build something that actually solves the problem you live with every day.