Definition
A company brain (also called an AI brain or AI knowledge base) is a company's shared, living memory: its policies, processes, decisions, and know-how, kept in one governed place that employees read and edit and that AI assistants search before they answer. It turns AI from a generic chatbot into a colleague who knows how your company works.
Why AI needs a company brain
AI assistants are fluent but uninformed. Ask one about your refund policy, your deploy process, or last quarter's pricing decision and it will either refuse or make something up. Teams work around this by pasting documents into chats, which creates three problems:
- Drift: everyone pastes a different, often outdated copy, so answers disagree.
- Waste: the same context is gathered again for every conversation.
- Leaks: sensitive documents end up in tools and chats no one governs.
A company brain solves all three: one current source that every AI tool reads from, with the same permissions people have.
What belongs in it
Start with the questions people ask most often:
- Handbook and policies: benefits, expenses, travel, security, refunds.
- How-to guides and runbooks: deploys, incidents, onboarding, month-end close.
- Decisions and their reasons: architecture decision records, pricing changes, meeting notes.
- Product and customer knowledge: specs, positioning, FAQs, troubleshooting.
Keep notes short and linked. A clear note titled "Refund policy" beats a 40-page document that mentions refunds on page 31.
How an AI knowledge base works
- Write: people keep knowledge in shared notes, ideally in a plain format like Markdown.
- Index: the knowledge base splits notes into passages and indexes them for search.
- Retrieve: when someone asks their AI a question, the assistant searches the knowledge base, through an API or the Model Context Protocol, and gets the most relevant passages, filtered by that person's permissions.
- Answer with citations: the assistant writes an answer from those passages and links to the source notes, so people can verify and fix them.
Retrieval vs fine-tuning
There are two ways to teach AI about your company. Fine-tuning retrains a model on your data: expensive, slow to update, hard to permission, and your data becomes part of a model. Retrieval (retrieval-augmented generation, or RAG) leaves the model alone and hands it the relevant passages at question time: always current, permission-aware, and reversible. For company knowledge, retrieval is almost always the right choice.
Five traits of a good company brain
- Easy to write in. If updating it is painful, it goes stale. Fast editing and live collaboration matter more than features.
- Permission-aware. AI should see only what the person asking may see.
- Cited. Every answer should point to its source.
- Model-neutral. You should be able to switch AI vendors without rebuilding your knowledge.
- Secure and portable. Encrypted, audited, never used for training, and exportable as plain files.
Granite is built around these five traits: Markdown notes with live co-editing, permission-filtered search with citations, a built-in MCP server for any AI tool, and encryption with a key per workspace. See how Granite's AI knowledge base works.
Frequently asked questions
What is the difference between a company brain and a wiki?
A wiki is where people write down company knowledge. A company brain is a wiki that AI tools can also use: searchable by assistants, permission-aware, and cited, so the same knowledge serves both people and AI.
What is an AI brain for a business?
An AI brain for a business is a central knowledge base that AI assistants and agents draw on to answer questions and do work with the company's real context, rather than generic training data.
Do we need a vector database to build a company brain?
Not if you use a knowledge base that handles indexing and retrieval for you. Granite indexes your notes and exposes search through MCP and an API, so there is no separate database or pipeline to run.
