Knowledge is not a store of facts. It is a structure of relationships, and it should be built like one.

Premise

A practice built on a single premise: what a person or an organisation knows is a living system of connections, and it should be kept as one.

Over a career, a person's thinking accumulates in conversations, drafts, transcripts, letters, and now a growing body of work done with AI. Over decades, an organisation's expertise accumulates in the people who hold it. In both cases the pieces are scattered and the connections between them are held in memory. Search finds documents; it does not find the reasoning between them. AI assistants are capable in the moment and forgetful the next day, so each conversation begins from nothing. When experienced people leave, what they could not write down leaves with them.

We treat knowledge as it actually is: relational, provenanced, changing over time, and inseparable from the conversations that produced it. The systems we build make those connections durable, inspectable and improvable, and record not only what is known but how it came to be known.

We claim no automation of judgement. No system decides what a thing means; a person does. Where others index documents, we keep the meaning, and the person in it.

What you would have

For a person, a memory that thinks with you. For an organisation, a collective intelligence that does not leave when people do.

As a person

You speak into your phone after a meeting, or share a link, or export a conversation you had with an AI. It becomes a record you keep. The system reads it, places it among your projects, and proposes what it contained — the ideas, the decisions, the questions left open — and shows you what it understood before it writes anything down. You approve, correct or set it aside, from wherever you are. Afterwards you can ask "what did I decide about this, and where?" and be answered with the conversation it came from. Over months the system notices what you keep returning to, and what two of your projects have in common that neither has named. Kept over years, it becomes a longitudinal graph of a life: experiences, relationships, ideas and artifacts, connected, with the paths by which each came to matter still visible.

As an organisation

An organisation does not think in one head. Its intelligence is collective, and it appears when the right people are brought into the right kind of conversation — a Syntegration when a whole system must deliberate on its own future, a Structured Democratic Dialogue when a group must structure its own observations, an Open Space when the agenda must come from the participants, a World Café when many small conversations need to cross-pollinate around questions that matter. We design and run the conversation the situation requires. What it produces — the observations, the structure the group found among them, the decisions and the reasons — enters the organisation's knowledge system as records with their provenance, reviewed by the people who made them. The next conversation starts from what the last one found. That is a collective intelligence with a memory, rather than a series of workshops whose output sits in a report.

Architecture

Every system we build has three parts, kept distinct: what it means, who does the work, and which AI happens to do it.

A graph that holds meaning

People, projects, ideas, claims, questions, decisions — and above all the relationships between them, each carrying where it came from and when. Documents stay outside the graph, as files the owner keeps. The graph is an index of meaning over them and can be rebuilt from the records at any time. It also keeps kinds distinct: evidence, interpretation, hypothesis and metaphor are marked as what they are, so that many people and many AIs can explore the same graph without a conjecture hardening into a fact or a metaphor being mistaken for evidence.

Agents organised like an organisation

Rather than a pipeline that runs the same steps every time, the agents hold roles and accountabilities and can question one another. They work in three layers, each answering a different question and each learning from the one below it. An operational layer asks what happened. A knowledge layer asks what does it mean. A meta-learning layer asks how is the system changing. Above the operational work sits a supervisory function that observes the observers.

A registry of capabilities

The system asks for capabilities — transcribe, propose links, reason over what is known — and a registry decides which provider supplies each: a cloud model, a local model on the owner's own hardware, a specialist tool. It also decides which material may go where. Providers are swapped without touching the architecture. The models are implementations; they are not the system.

All three are run as one loop for every conversation, personal or collective: capture, propose, review, then reason and learn. The unit of record is the conversation, not the isolated fact. Nothing is deleted; it is superseded.

Services

Four ways we put the architecture to work.

A personal knowledge system — your own

A single person's graph of their own thinking across every role they hold — the writing, the projects, the correspondences, the conversations with colleagues and with AI. Captured from a phone, reviewed from anywhere, and able to answer "what did I decide about this, and where?" with its provenance. The records stay in the owner's private repository; the code that runs the system never needs to see them.

Collective intelligence for organisations

The design and facilitation of the conversation an organisation needs — Syntegration, Structured Democratic Dialogue, Open Space, World Café, or a combination chosen for the situation — with its outcomes captured into an organisational knowledge system the organisation owns. Governance to match: who may propose, who may approve, what counts as private, and which standards the system is held to. Personal and organisational graphs can be linked without being merged, so each keeps its own governance.

Expert knowledge, drawn out and made learnable

Expertise drawn out through structured conversation rather than documentation, organised so that others can learn from it, and understanding judged by whether a learner can explain and teach it back rather than recall it. This is the learning strand of the architecture, and it rests on fifty years of research into how understanding forms through dialogue.

Design and advisory

For organisations building their own: the ontology, the agent roles, the review gate, the governance model, and the boundary between what stays local and what may leave. We advise on the design, and on the standards and audit trail that let it be trusted.

Commitments

These are design decisions, not features. They are the difference between a store of information and a memory that thinks.

Foundation

The architecture is cybernetic in the proper sense: a system of regulating loops in which each layer observes and improves the one below it, and in which the person is not a user of the system but part of it. The learning strand rests on a theory of how understanding forms through conversation — how it differs from recall, and how to tell the two apart by asking someone to explain a thing back. The same test is applied to the agents before they are allowed to act.

The technology to implement this — conversational AI, graph databases, capable local models — has only recently become possible. The theory has been waiting for it for fifty years. More on the methodology.

Contact

To discuss a knowledge system of your own, brief us on what you are trying to hold together, or explore whether this architecture applies to your work, write directly. We respond to every enquiry, and every engagement is held in confidence.

peter.tuddenham@coexplorer.com

Northern Virginia, United States