> ## Documentation Index
> Fetch the complete documentation index at: https://help.iteam.works/llms.txt
> Use this file to discover all available pages before exploring further.

# Knowledge

> Your company's knowledge base inside Cortex: documents (uploaded or written right there) and a glossary of terms, organized by area. Everything stays searchable by the agents by meaning, respecting who can see what. It's iTeam's 'smart Notion'.

**Knowledge** is your company's base of documents and definitions, inside Cortex. Instead of knowledge being scattered across drives and in people's heads, you keep it here — and the **agents start answering based on it**. Think of a "smart Notion": beyond storing, it **understands** the content and brings it up at the right moment.

<Frame caption="📸 Screenshot to insert [/en/cortex/knowledge]: Knowledge page with the list of documents by area, the glossary of terms and the button to add a document; 'Ask Cortex' button at the top.">
  <img src="https://mintcdn.com/iteam/E1mg8rFlXblsaNT0/images/placeholder.svg?fit=max&auto=format&n=E1mg8rFlXblsaNT0&q=85&s=36c435f4df28c31a7e114586a931bbd2" alt="Knowledge base" width="800" height="420" data-path="images/placeholder.svg" />
</Frame>

## How it works, in one sentence

You add **documents** and **glossary terms** (by area) → iTeam **indexes** the content by meaning → when an agent (or Cortex) needs it, it **searches and uses** it automatically, respecting the **permissions** of whoever asked.

## Two things you keep here

<CardGroup cols={2}>
  <Card title="Documents" icon="file-lines">
    Playbooks, policies, procedures, business context. You can **upload a file** or **write the text right there** in the editor.
  </Card>

  <Card title="Glossary" icon="book-open">
    The **terms** and acronyms of your company/niche, with their definitions. E.g. in iGaming: *GGR, NGR, FTD, MTD, TDD*. That way the agent speaks your language.
  </Card>
</CardGroup>

## Add a document

<Steps>
  <Step title="Choose the area">
    Every document belongs to an **area** (e.g. Marketing, Finance). This organizes the base and controls who sees it.
  </Step>

  <Step title="Upload or write">
    **Upload a file** (iTeam extracts the text) **or** write directly in the built-in **text editor**. Give it a clear title and a description — they help the search find the document later.
  </Step>

  <Step title="Set the visibility">
    **Company** (everyone sees it) or **Area** (only those who belong to that area). Private area content doesn't show up for those outside the area — not even in the agents' answers.
  </Step>
</Steps>

<Info>
  On saving, the document is **indexed** (turned into a representation the AI searches by meaning). Large documents may take a few moments to become 100% searchable — the title and description are available right away.
</Info>

## How the agents use it

The content doesn't sit idle: when you ask an agent (or **Cortex**) something, it **searches the knowledge base** and uses the relevant passages to answer — citing what it found. That's why it's worth feeding the base: each document makes **all the agents** better prepared.

<Tip>
  Start with your **niche's glossary** and **2 or 3 key documents** (a support policy, business rules, a playbook). The payoff shows up in the very first conversation with an agent.
</Tip>

## Search by meaning

The search is **semantic**: you don't need to hit the exact word. Asking *"what's our discount limit?"* finds the right passage even if the document says *"commercial concession policy"*.

## Permissions and privacy

Knowledge respects access by **resource** and by **area**:

* Documents with **Company** visibility — all members see them.
* Documents with **Area** visibility — only those who belong to the area (owner/admin see everything).
* The agents and Cortex **never** bring, in an answer, content the asking user couldn't see.

<Warning>
  Privacy also applies to the AI's answers: if a document is private to an area, it's **not** considered when the person asking isn't from that area. This is guaranteed in the search mechanism itself — it's not just "hiding it on screen".
</Warning>