kbrain

Architecture

The anatomy of a brain: what is inside a KBrain knowledge brain

A brain is more than a pile of documents. It has sources, metadata, instructions, permissions, a domain, an index, and an MCP endpoint. Here is what each part does and why it matters.

Build your first knowledge brain

Create a brain

A brain is not just a folder of files. It is a curated knowledge asset with distinct parts, each doing a specific job - from the raw sources it is built on, to the metadata that tells an assistant when to use it, to the MCP endpoint that makes it queryable. Understanding the anatomy makes it much easier to build brains that work.

The parts of a brain

Anatomy - 01
What is inside a brain
Each part does one job. Together they turn documents into answerable knowledge.
🗂️
Sources
the raw material

Files and PDFs, Google Drive, GitHub, URLs, and text notes - the documents the brain is built from.
🏷️
Metadata & domain
what and when

Title, description, category, and intents. This is how an assistant decides the brain is the right one to query.
📋
Instructions
how to answer

Guidance on tone, scope, and how the brain should be used when an assistant draws on it.
🔐
Permissions
who can query

Visibility - private, public in the marketplace, a shareable open link, or shared org-wide.
🔎
Index
made retrievable

Sources are processed so the right passage can be retrieved at query time, not the whole pile.
📡
MCP exposure
made queryable

The brain is served over MCP, so Claude, ChatGPT, and any MCP client can query it.
The sources are the body. The metadata is the label. The MCP endpoint is the door. A good brain gets all three right.

Sources: what the brain knows

Sources are the documents a brain is built from - uploaded files and PDFs, a connected Google Drive folder, a GitHub repository, URLs, or short text notes. The quality of a brain starts here: curated, relevant sources make a focused brain; everything dumped in makes a noisy one.

Metadata and domain: when to use it

Metadata - the title, description, category, and intents - is what an assistant reads to decide whether this brain is the right one for a question. It is not decoration. When you have several brains, the description is the routing signal that sends a request to the right one.

Permissions and MCP exposure: who and how

Every brain has a visibility setting that controls who can query it, and an MCP endpoint that makes it queryable at all. Together they decide reach: a private brain only you can query, a public brain anyone can subscribe to, an open link you can share, or a brain shared across everyone in your organisation.

Build your first knowledge brain

Subscribe to KBrain, create a brain from your expertise or your data, and make it available to Claude, ChatGPT, or any MCP compatible assistant.

Create a brain

Frequently asked questions

What is inside a KBrain brain?

A brain has sources (files, Google Drive, GitHub, URLs, notes), metadata (title, description, category, intents), instructions on how it should be used, a visibility setting for permissions, an index that makes it retrievable, and an MCP endpoint that makes it queryable by any assistant.

Why does the brain description matter so much?

The description is metadata an assistant reads to decide whether this brain answers a given question. When you have more than one brain, the description is the routing signal that sends a request to the right brain, so a clear, specific description directly improves answer quality.

How big should a brain be?

Focused beats large. A brain should cover one domain or workflow well rather than everything at once. A tightly scoped brain is easier for an assistant to route to and produces more reliable answers than a monolithic knowledge base.