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

# RAG (Knowledge Base)

> Use Burki RAG knowledge bases to upload documents, index custom content, and help voice AI assistants answer from approved sources.

<Callout type="info">
  RAG (Retrieval-Augmented Generation) lets you give your assistant custom knowledge by uploading documents, FAQs, and other content. Your AI can then reference this information to give accurate, specific answers.
</Callout>

***

## What is RAG?

**RAG** combines your custom knowledge with the AI's general intelligence:

1. **Upload documents** (PDFs, text files, web pages, etc.)
2. **AI processes and indexes** the content
3. **During calls**, the AI searches your knowledge base
4. **Combines** retrieved information with its response

<Callout type="success">
  Think of RAG as giving your AI a "company handbook" it can instantly reference during conversations.
</Callout>

***

## How to Set Up RAG

<Accordion title="Step 1: Prepare Your Content">
  **Supported Formats:**

  * PDF documents
  * Text files (.txt, .md)
  * Web pages (URLs)
  * Structured data (JSON, CSV)

  **Content Tips:**

  * Keep documents focused and well-organized
  * Use clear headings and sections
  * Include FAQs, product info, policies, etc.
  * Remove outdated or irrelevant information
</Accordion>

<Accordion title="Step 2: Upload to Your Assistant">
  1. Go to your assistant's **RAG Configuration** section
  2. Click **"Upload Documents"** or **"Add Knowledge"**
  3. Upload files or paste URLs
  4. Wait for processing (may take a few minutes)
  5. Test with sample questions

  <Callout type="tip">
    Start with your most important documents—you can always add more later!
  </Callout>
</Accordion>

<Accordion title="Step 3: Configure RAG Settings">
  * **Similarity Threshold:** How closely questions must match content
  * **Max Results:** Number of knowledge pieces to retrieve
  * **Context Window:** How much text to include in responses

  <Callout type="warning">
    Higher similarity thresholds are more selective but may miss relevant content. Start with default settings and adjust based on testing.
  </Callout>
</Accordion>

***

## Use Cases

<Accordion title="Customer Support">
  * Product manuals and troubleshooting guides
  * Company policies and procedures
  * FAQ documents
  * Return/refund policies

  **Example:** "How do I return a product?" → AI finds your return policy and gives specific instructions.
</Accordion>

<Accordion title="Sales & Information">
  * Product catalogs and specifications
  * Pricing information
  * Company background and history
  * Service descriptions

  **Example:** "What services do you offer?" → AI references your service catalog for accurate details.
</Accordion>

<Accordion title="Internal Knowledge">
  * Employee handbooks
  * Internal procedures
  * Technical documentation
  * Training materials

  **Example:** "What's our vacation policy?" → AI finds the relevant HR policy section.
</Accordion>

***

## Best Practices

* **Quality over quantity:** Better to have fewer, high-quality documents than many poor ones
* **Keep content updated:** Regularly review and update your knowledge base
* **Test thoroughly:** Ask questions you expect callers to ask and verify accuracy
* **Use clear language:** Write content in plain language, avoiding jargon when possible
* **Organize logically:** Group related information together

***

## Troubleshooting

<Accordion title="Common RAG Issues">
  * **AI doesn't find relevant info:** Check similarity threshold, add more specific keywords
  * **Outdated information:** Remove or update old documents
  * **Conflicting answers:** Ensure documents don't contradict each other
  * **Too much irrelevant content:** Use more focused, specific documents
</Accordion>

<Callout type="tip">
  RAG works best when combined with a well-written system prompt that tells the AI how to use the knowledge base effectively.
</Callout>