Governance

Designing high-quality brains: scope, sources, and best practices

A great brain is focused, well-named, and built from curated sources. Here are the best practices for scope, naming, metadata, instructions, and sources - and the common mistakes to avoid.

Build your first knowledge brain

Create a brain

Two brains built from the same documents can produce very different results. The difference is design: how tightly the brain is scoped, how clearly it is named and described, and how carefully its sources are curated. This is a practical guide to building brains an assistant answers well from.

Focused beats broad

The core choice - 01

A focused brain vs. a monolith

Scope is the highest-leverage design decision you make.

🗃️Monolithic brain

· Everything in one place, no clear boundary

· Vague description an assistant cannot route on

· The right fact competes with a hundred wrong ones

· No single owner, so it goes stale

hard to route
hard to trust

🎯Focused brain

→ One domain or workflow, clear boundary

→ Specific description that routes cleanly

→ Curated sources, less noise

→ A clear owner who keeps it current

✓

routes well
answers well

When in doubt, split it. Two focused brains almost always beat one that tries to cover both.

Best practices

Scope: one domain or workflow per brain - if you cannot describe it in a sentence, it is too broad
Naming: name the brain for what it knows, not for the team that owns it
Description: write it for an assistant deciding whether to use the brain - be specific about what it covers
Instructions: say how the brain should be used, including tone and what it should not answer
Sources: curate. Add the material that matches the audience and workflow, not everything you have

Common mistakes

Dumping every document in and hoping retrieval sorts it out
A vague description like "company info" that gives an assistant nothing to route on
Mixing audiences - internal and customer-facing knowledge in the same brain
Building it once and never updating the sources

The description is the highest-return thing to get right. It is what an assistant reads to decide whether to query the brain at all. Spend more time on it than feels necessary.

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 makes a high-quality brain?

A tight scope (one domain or workflow), a specific description an assistant can route on, clear instructions, and curated sources that match the audience. Focused, well-described brains produce far more reliable answers than large, vague ones.

How specific should a brain description be?

Very. The description is what an assistant reads to decide whether to query the brain, so it should clearly state the domain and the kinds of questions the brain answers. A vague description like "company info" gives the assistant nothing to route on.

What is the most common mistake when building brains?

Dumping in every document and hoping retrieval sorts it out. Curation is what makes a brain sharp - add the material relevant to the brain’s audience and workflow, and split anything that spans two distinct domains into separate brains.