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

# Knowledge Base

> How Parlel uses your content to ensure the AI always answers accurately — never from generic training data.

## Overview

The Knowledge Base is how Parlel ensures the AI speaks with your voice and your facts. Every answer the AI gives to a customer is grounded in content you've configured — your products, policies, pricing, and procedures. The AI never falls back to generic training data when answering questions about your company.

This is the foundation everything else builds on. Without a well-configured knowledge base, the AI has nothing authoritative to draw from.

## How it works

When a customer sends a message, Parlel searches your knowledge base for the most relevant content using semantic retrieval — matching meaning, not just keywords. The most relevant content is provided to the AI as the authoritative source for its response.

If the AI can't find relevant content with sufficient confidence, it doesn't guess. Instead it escalates cleanly — either acknowledging the limitation or triggering a [Fallback Flow](/fallbacks/fallback-flows) you've configured.

This means customers always get accurate answers or a clear handoff. Never a confident wrong answer.

## Relationships

<CardGroup cols={2}>
  <Card title="Fallback Flows" icon="life-ring" href="/fallbacks/fallback-flows">
    When knowledge retrieval confidence is low, Fallback Flows define what happens next — escalation to an agent, a redirect, or a configured response.
  </Card>

  <Card title="Actions" icon="bolt" href="/actions/actions">
    Knowledge and actions complement each other. Knowledge handles questions; actions handle requests that require changes to customer data.
  </Card>
</CardGroup>

## Configuration

### Adding content

Go to **Knowledge** in the portal. Parlel accepts content in the following formats:

* PDF documents
* Word documents (.docx)
* Plain text and Markdown
* Web URLs (Parlel fetches and indexes the content)

Upload your content and Parlel processes it automatically — splitting it into retrievable chunks, generating embeddings, and indexing it for semantic search.

### Organising content

Content is organised into **knowledge blocks** — logical groupings that help Parlel retrieve the right content for the right question. Good organisation improves retrieval accuracy.

Recommended blocks for most companies:

* Product catalogue and descriptions
* Pricing and plans
* Policies (cancellation, returns, terms)
* FAQs and support articles
* Onboarding and how-to guides

### Keeping content current

When your content changes, update or replace the relevant knowledge blocks in the portal. Parlel reprocesses updated content automatically — changes are available to the AI immediately after processing.

## Reference

| Concept              | Description                                                            |
| -------------------- | ---------------------------------------------------------------------- |
| Knowledge block      | A logical grouping of related content                                  |
| Chunk                | A retrievable segment of a knowledge block                             |
| Semantic retrieval   | Matching by meaning, not exact keywords                                |
| Confidence threshold | Minimum confidence score required before the AI uses a retrieved chunk |
| Escalation           | What happens when no chunk meets the confidence threshold              |