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

# Give Your AI Voice Agents Accurate Business Knowledge

> Upload documents or crawl your website to give VoiceInfra agents accurate, real-time knowledge. Retrieval under 100ms with AES-256 encryption.

The VoiceInfra Knowledge Base lets you give your AI agents access to your actual business content, product manuals, pricing sheets, FAQ documents, support articles, or your entire website, so they answer questions with documented facts instead of guesses. When a caller asks something specific, the agent retrieves the relevant content from your knowledge base in under 100ms and incorporates it directly into the response. No hallucinations. No outdated answers. No hardcoded scripts that someone has to maintain manually.

## Adding Content

You can populate a knowledge base two ways: by uploading documents directly or by crawling a website URL.

### Upload Documents

<Steps>
  <Step title="Open the Knowledge Base dashboard">
    Navigate to **Knowledge Base** in the left sidebar. The dashboard lists all your existing sources with their processing status: **Completed**, **Review Pending**, or **Failed**.

    <Frame>
      <img src="https://mintcdn.com/voiceinfraai/-BxpdbvZpQEOPuHc/images/image-11.png?fit=max&auto=format&n=-BxpdbvZpQEOPuHc&q=85&s=c19b1cf57522e1b3f615eb3df19dc56c" alt="Image" width="1859" height="860" data-path="images/image-11.png" />
    </Frame>
  </Step>

  <Step title="Upload your files">
    Click **Add Content** and select **Upload File**. Drag and drop your files or browse to select them. You can upload multiple files at once.

    Supported formats: PDF, DOCX, DOC, TXT, up to **10 MB per file**. There is no limit on the number of documents you can upload.
  </Step>

  <Step title="Wait for processing">
    The platform extracts and indexes the content automatically. Most documents complete in under a minute. The status updates to **Completed** when the document is ready.
  </Step>

  <Step title="Attach to an agent">
    Open any agent's configuration and navigate to the **Knowledge Base** tab. Select the documents you want that agent to reference. One document can be attached to as many agents as you need.
  </Step>
</Steps>

### Crawl a Website

Enter any public URL and VoiceInfra fetches, parses, and indexes the content. Choose the crawl depth that matches how much of the site you want to capture.

| Crawl depth  | Pages crawled                                              | Typical duration |
| ------------ | ---------------------------------------------------------- | ---------------- |
| **Quick**    | Single page only                                           | 30 – 60 seconds  |
| **Standard** | Starting page + 1 level of linked pages (10 – 50 pages)    | 2 – 5 minutes    |
| **Deep**     | Starting page + 2 levels of linked pages (50 – 200+ pages) | 5 – 10 minutes   |

After the crawl completes, VoiceInfra enters a **Review Pending** state before committing the content to your knowledge base:

1. **Review extracted pages** — see every page title, URL, and full content preview side by side.
2. **Edit content** — fix formatting issues, correct extracted text, or add context the crawler may have missed.
3. **Discard irrelevant pages** — exclude pages you don't want the agent to reference, such as blog posts, legal boilerplate, or unrelated sections.
4. **Confirm** — publish only the pages you've approved. The agent never sees content you discarded.

<Frame>
  <img src="https://mintcdn.com/voiceinfraai/9FH9hUvZZvUOANCA/images/image-12.png?fit=max&auto=format&n=9FH9hUvZZvUOANCA&q=85&s=3846407d254f8d123dd89912418a41c7" alt="Image" title="Image" className="mx-auto" width="1768" height="547" data-path="images/image-12.png" />
</Frame>

## How Retrieval Works

When a caller asks a question that requires business-specific knowledge, the agent issues a semantic search query against your knowledge base in real time. Semantic search matches meaning, not just keywords, so a caller asking *"what's your refund policy?"* matches a document section titled *"Returns and Exchanges"* even though the words differ.

* **Retrieval latency** — under 100ms in production, so conversations stay natural with no perceptible pause
* **Concurrent scale** — retrieval is optimized for high-volume simultaneous calls

## Sharing Across Agents

A single knowledge base document can be attached to unlimited agents simultaneously. This means you can upload your product catalog once and share it across your sales agent, your support agent, your website widget, and your WhatsApp agent, all from one source of truth.

## Technical Specifications

| Specification              | Detail                                                       |
| -------------------------- | ------------------------------------------------------------ |
| **Supported file formats** | PDF, DOCX, DOC, TXT                                          |
| **Maximum file size**      | 10 MB per file                                               |
| **Document limit**         | No limit                                                     |
| **Crawl depths**           | Quick (1 page), Standard (10–50 pages), Deep (50–200+ pages) |
| **Retrieval latency**      | Under 100ms                                                  |
| **Encryption**             | AES-256 at rest and in transit                               |
| **Uptime SLA**             | 99.9%                                                        |
| **Multi-agent sharing**    | Yes — one document, unlimited agents                         |

***

## Frequently Asked Questions

<AccordionGroup>
  <Accordion title="What file formats are supported?">
    You can upload PDF, DOCX (Microsoft Word 2007 and later), DOC (older Word formats), and TXT (plain text) files. Each file can be up to 10 MB. There is no limit on the number of documents you can add to a knowledge base.
  </Accordion>

  <Accordion title="Can I edit crawled content?">
    Yes. After a crawl finishes, VoiceInfra puts the content in **Review Pending** status before it goes live. You can read through every extracted page, edit the text directly in the review interface, and discard any pages you don't want included. Only the content you explicitly confirm gets published to the knowledge base.
  </Accordion>

  <Accordion title="Can multiple agents share the same document?">
    Yes. Attach a single document to as many agents as you need.
  </Accordion>

  <Accordion title="How long does crawling take?">
    Crawl duration depends on the depth you choose and the size of the target website:

    * **Quick** (single page): 30 – 60 seconds
    * **Standard** (page + 1 level, 10–50 pages): 2 – 5 minutes
    * **Deep** (page + 2 levels, 50–200+ pages): 5 – 10 minutes for large sites

    The status indicator on the dashboard updates in real-time so you can track progress.
  </Accordion>
</AccordionGroup>