Preface Executive Briefing

Week of July 18, 2026

From database to knowledge base

The missing infrastructure behind agentic transformation.

The gap

More AI, and still no lift

Most organisations now have more AI than ever, and little to show for it. The assistants are capable. What they lack is a shared, current picture of how your organisation actually works.

AI assistants

ChatGPT

Gemini

Claude

No shared context

Your files and data

strategy-memo.pdf

customer-feedback.csv

monthly-pnl.xlsx

An AI Knowledge Base is that shared context. It is your organisation's world written down as linked pages the AI can read, add to and keep current, so every assistant works from the same durable memory instead of starting cold each time.

Why it slips

Three reasons knowledge never sticks

Today most of what an organisation knows lives in one of three forms, and each one leaks.

Chat history is fragile

It gets buried, is hard to retrieve, and is rarely saved.
The assistant forgets the earlier parts of the conversation.

Documents are disconnected

Files sit in folders, never linked or updated.
The AI cannot see how one file relates to another.

Judgement is scattered

Decisions, principles and rationale live in many places.
Tone, style and preferences exist only in your head.

What's missingA durable, structured memory the AI can understand, maintain and use, not a transcript that scrolls away or a folder it cannot read.

The knowledge point

Tools let it act. Context tells it what matters.

An agent can already see, plan and do. It reaches the web, your files, your inbox. What sits between those actions and everything your organisation knows is the layer most teams are missing.

Your agent

Observe

Plan

Act

Tools

webfilesemail

AI Knowledge Base

The missing layer between what the agent does and what your organisation knows.

Your files and data

Policies

Reports

Decisions

Agents rarely fail because they lack tools. They fail because they lack context.

The idea

Your organisation as a living wiki

An AI Knowledge Base turns the people, sources, concepts, projects and decisions your organisation runs on into linked pages the agent builds, connects and keeps current.

Ask it anything. It answers from your business context, and cites the pages it used.

An AI Knowledge Base shown in Obsidian: a folder tree of entities, concepts and sources beside a linked graph view of hundreds of connected pages.

One real example: a single executive's world, organised into a living wiki the agent keeps current.

111

entities

concepts

sources

300+

linked pages

Two layers

Knowledge gives facts. Identity gives judgement.

A knowledge base has two parts. One holds what your organisation knows. The other holds how it decides.

Knowledge layer

What it knows

The facts: people, companies, concepts and sources, all linked together. The shared record everyone can draw on.

Identity layer

How it decides

The judgement: how a person or a team decides, what they value, how they speak. The part that usually lives only in someone's head.

Knowledge+identity=a digital replica

Wiki vs database

Why a wiki beats a database

A database gives you back exactly the rows you ask for. A knowledge base lets you ask open questions, and it keeps getting better at answering them.

“Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge.”

Andrej Karpathy, Co-Founder, OpenAI

Retrieval, every time

It starts from scratch every time

The usual approach fetches snippets and works the answer out again on every question. Nothing is kept, so the effort repeats.

A living wiki

It writes the answer down and compounds

The wiki records what it worked out once, links it to the rest, and improves it over time. Each question builds on the last instead of restarting.

The pieces

What it's actually made of

Two parts do most of the work. One acts on your files. The other is how you see and steer what it writes.

The agent

It works with your files, not just chat

A chatbot only talks back. A file-acting agent opens your files, reads them and writes new ones, so it can build and maintain the base itself.

Runs on your machine, so your files stay yours.
Use whichever AI model you prefer, including your own.

Any agent that can reach your files works.

Claude CodeOpenCodeCodex

The workspace

Obsidian, your window into the files

The agent writes plain Markdown, a type of text file. Obsidian shows those files as a linked graph you can read, search and steer.

Ordinary text files you can open anywhere.
Links between pages become the graph.
No database and no lock-in: just files you own.

How it runs

Ingest, link, ask

Day to day, the base runs on a simple loop, driven by repeatable jobs you trigger on command.

Ingest

Hand it a source: a document, an email thread, a meeting note.

Link

It files what matters into connected pages, joined to what it already knows.

Ask

Query the base in plain language and get an answer that cites its pages.

Skills

Repeatable jobs, run the same way each time

A skill is a job the agent runs on command, so building and using the base stays consistent.

ingestquerylintidentity

Connectors (MCP)

Reach into the tools you already use

A connector links the agent to your existing tools without leaving the folder, so it can pull in what it needs.

Google Drivethe webNotion

Shaping it

A base is only as good as how you shape it

The tools are the easy part. The value comes from what you choose to put in, and keeping it honest.

Start with the entities that matter. The people, companies and products you actually deal with.

Capture the concepts. The ideas your work turns on, and how your organisation uses them.

Keep the sources. So every answer can point back to where it came from.

Write down decisions. The reasoning, not just the outcome, so judgement is preserved.

Keep it current. A knowledge base is a living thing. Its value comes from being shaped and maintained, not from being set up once and left.

Key Takeaways

Context is the infrastructure, not the model

The advantage in agentic AI is shifting from which model you use to the durable, structured context you give it.

The bottleneck is context, not tools

Buying more assistants adds capability the AI cannot aim. The durable advantage is a shared, structured memory it can read and maintain, so the work compounds instead of restarting in every new chat.

A knowledge base is a wiki, not a database

Linked pages the agent builds and keeps current let you ask open questions a table never could, and get answers that cite their sources. It grows more useful over time rather than starting from scratch on every question.

You own the shape and the files

It is plain Markdown you can open anywhere, steered through Obsidian and run by repeatable skills. There is no lock-in, and the base is only as good as the entities, sources and decisions you choose to put in it.

FAQ

It is your organisation's world written down as linked pages an AI can read, add to and keep current. Instead of scattered files and chat histories, the people, sources, concepts and decisions you run on live in one connected place the agent works from.