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

# Knowledge base

> Brand Model plus company files — the constitution Agents follow when they write.

**Knowledge base** is where you tell UX Report who the company is and which
source documents Agents may retrieve. It sits **above Settings** in the
sidebar — not inside Analytics or Optimize.

## Two streams (they stay separate)

| Stream            | What it is                                                                                                  | How Agents use it                                                                |
| ----------------- | ----------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------- |
| **Brand Model**   | Structured fields: summary, ICP, audiences, goals, objections, benefits, writing style, tone, writing rules | Always applied to content generation. Missing Brand Model blocks content drafts. |
| **Company files** | Uploaded or pasted documents (product, messaging, pricing, FAQ, …)                                          | Retrieved by default for content jobs. Empty corpus **warns** and still runs.    |

Uploaded files are **never** merged into Brand Model fields. Edit the Brand
Model when identity or voice changes; add files when you want extra evidence.

Visibility checks (engine probes, Prompt Factory) use Brand Model only — they
do not retrieve company files.

## Brand Model

Open **Knowledge base** and edit the Brand Model section. Most fields are
free text. **Audience** is a name + description list. An optional writing-sample
URL can refresh style from a live page.

First-run infers a draft from your homepage in the background. You do not have
to finish a brief wizard before Monitoring. Come back here to correct it.

## Company files

Drop or attach text documents. Accepted formats today are text-first
(`.md`, `.txt`, `.csv`, `.json`, `.html`, `.yaml`, and similar) — not PDF or
Office binaries yet.

Each file gets a kind (Product, Messaging, Pricing, FAQ, Policies, Technical,
Competitive, General) so retrieval stays scoped.

## Typical flow

1. After your first monitoring pass, open **Knowledge base**.
2. Correct Brand Model so claims, ICP, and tone match how you actually speak.
3. Drop the docs Agents should be allowed to quote (product sheets, approved FAQ, pricing facts).
4. Run an [Agent](/product/agents) template — drafts should follow that constitution.

## Related

* [Agents](/product/agents)
* [Opportunities](/product/improve)
* [Getting started](/getting-started)
* [Product overview](/product/overview)