kbrain

Enterprise architecture

Building an enterprise AI knowledge stack

An enterprise AI knowledge stack has four layers: assistants, MCP, a knowledge layer, and your sources. Here is how they fit together, with a simple architecture diagram.

Build your knowledge layer

Connect via MCP

Every enterprise rolling out AI ends up assembling the same stack, whether they plan it or not. There are four layers: the assistants people use, the MCP protocol that connects them, a knowledge layer that curates what those assistants know, and the sources the knowledge is built from. Naming the layers makes the architecture obvious - and shows where KBrain fits.

The four layers

Architecture - 01
An enterprise AI knowledge stack
Four layers, each doing one job. MCP is the seam that holds them together.
1 · Assistants
Claude, ChatGPT, and any MCP-compatible agent your teams already use.
↕  MCP  ·  the standard interface  ↕
2 · MCP
One open protocol carries every request. Connect once, reach every brain.
3 · Knowledge layer KBRAIN
Curated brains with owners, visibility (private / public / open / organization), and metadata that routes each question to the right brain.
4 · Sources
Documents and PDFs, Google Drive, GitHub, and URLs - brought into brains via upload or sync.
The model reasons. MCP connects. The knowledge layer curates. Your sources supply the truth. Miss the knowledge layer and assistants are left guessing from training data.

Where your existing systems fit

Your documents live in the tools you already use - Google Drive, a wiki, a repo, a document store. In the stack, those are the source layer. You bring their content into brains by connecting a Google Drive folder or a GitHub repository, or by uploading files and adding URLs. The knowledge layer curates that material; it does not replace the systems it came from.

Why the knowledge layer is the piece teams miss

Most enterprises already have the assistants and the sources. What they are missing is the layer in between - the one that decides what an assistant should know, keeps it current, and controls who can reach it. Without it, every team improvises with pasted context and one-off connectors. The knowledge layer is what turns that into an architecture.

You do not have to build the knowledge layer from scratch. KBrain is that layer: curated brains, org-wide sharing, access control, and one MCP endpoint - without hosting infrastructure.

Build your knowledge layer

Put a curated knowledge layer between your assistants and your sources. Create a brain, connect your documents, and expose it over MCP.

Connect via MCP

Frequently asked questions

What is an enterprise AI knowledge stack?

It is the four layers that connect AI to your knowledge: the assistants people use (Claude, ChatGPT), the MCP protocol that connects them, a knowledge layer that curates what they know, and the sources that knowledge is built from. Each layer does one job.

Where does KBrain sit in the stack?

KBrain is the knowledge layer - between the assistants and your sources. It holds curated brains with owners, visibility controls, and routing metadata, and exposes them over MCP so any assistant can query them. It complements your systems of record rather than replacing them.

Do I connect my existing tools directly?

You bring their content into brains. Connect a Google Drive folder or a GitHub repository, or upload files and add URLs. Those sources feed the knowledge layer, which curates and exposes the material to your assistants over MCP.

Which layer do most companies already have, and which is missing?

Most already have the assistants and the sources. The missing piece is the knowledge layer in between - the part that decides what assistants should know, keeps it current, and controls access. That is the gap KBrain fills.