> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://poly-ai.ferndocs.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://poly-ai.ferndocs.com/_mcp/server.

# Sources

> Manage multiple external knowledge sources (sources) to inform your agent's responses.

Import existing content – help articles, PDFs, internal docs – so your agent can reference it without rewriting everything as individual topics. Connected Knowledge aggregates sources and re-syncs automatically.

![collected-knowledge](/_fern-files/_img/7edd81eddd0ac75e50caa87b728c47da97c47925e9107239d05a97352ba31d9f.webp)

The **Connected** tab is found under **Knowledge > Sources** in Agent Studio. [Raven](/agent-studio/behavior/models/overview) is the recommended model — it paraphrases unstructured content more naturally than other models.

> **Note**
>
> **Use Connected Knowledge when** you want to expose large volumes of external content quickly without curating individual topics. **Use [FAQs](/agent-studio/knowledge/fa-qs/fa-qs) instead** when you need actions, flows, or precise control over what the agent says and does. Both use [RAG (retrieval-augmented generation)](https://en.wikipedia.org/wiki/Retrieval-augmented_generation) to match user queries.

## Supported sources

* Websites
* Documents (PDF, CSV, JSON)
* Help desk systems (Zendesk, Gladly)

Sources sync automatically and can be reused across projects.

## How Sources differs from FAQs

Both tabs expose information to your agent. Key differences:

| Capability                              | Connected tab                     | FAQs tab                                         |
| --------------------------------------- | --------------------------------- | ------------------------------------------------ |
| Trigger actions, functions, flows, SMS  | No                                | Yes                                              |
| Precise control over agent responses    | No                                | Yes                                              |
| Auto-sync from external sources         | Yes                               | No                                               |
| Best for frequently updated FAQ content | Yes                               | --                                               |
| Best for stable, structured info        | --                                | Yes                                              |
| Fine-grained behavior control           | No                                | Yes                                              |
| Setup complexity                        | Low – no prompting skill required | Higher – requires more expertise and maintenance |

**Connected** = fast import of external content. **FAQs** = precise control with actions and flows.

If both tabs contain conflicting information, **FAQs always takes priority**.

## Add a new source

1. Go to **Knowledge > Sources tab**
2. Select **New source**
3. Choose one of:
   * **Upload files**
   * **Add URL**
   * **Zendesk**
   * **Gladly**
   * Additional integrations are in development – contact your PolyAI representative for the latest availability
4. Complete the required details and click **Add**

![new-source](/_fern-files/_img/165d142b28d9658f96ab06c9b90bc1d2d3a3c31072a05055946e65b60f3cdb6b.webp)

Your agent will begin **Syncing** the content. Once ready, the source appears in the list.

## Supported source types

| Source Type                               | Details                                                                              |
| ----------------------------------------- | ------------------------------------------------------------------------------------ |
| **Upload files – Text & structured data** | `.txt`, `.csv`, `.json`, `.xml`, `.md`, `.html`, `.rtf`                              |
| **Upload files – PDF**                    | `.pdf`                                                                               |
| **Upload files – Microsoft Office**       | `.docx`, `.doc`, `.docm`, `.xlsx`, `.xls`, `.xlsm`, `.pptx`, `.ppt`, `.pptm`, `.msg` |
| **Upload files – OpenDocument**           | `.odt`, `.ods`, `.odp`                                                               |
| **Upload files – Email files**            | `.eml`                                                                               |
| **Upload files – E-books**                | `.epub`                                                                              |
| **URL scraping**                          | Public documentation pages and help center articles                                  |
| **Zendesk** *(beta)*                      | Help Center content with API sync                                                    |
| **Gladly** *(beta)*                       | Knowledge source sync                                                                |
| Additional integrations                   | In development – contact your PolyAI representative for the latest availability      |

## What exactly gets scraped when I upload a URL?

URL scraping traverses linked pages from the provided URL, with the following limits:

1. **Depth** → Only one level below the initial [URL](https://en.wikipedia.org/wiki/URL).
2. **Breadth** → A maximum of 10 embedded pages.

If your page contains more than 10 links, not all will be scraped. In that case, upload additional URLs individually or use integrations like [Zendesk](/integrations/crm/zendesk-talk)/[Gladly](/integrations/knowledge-base/gladly) for complete coverage.

Where possible, connect applications such as Zendesk rather than relying on website scraping.

## Keeping content fresh

After external content changes:

* click **Update** to re-scrape files or URLs
* or use the **Sync** icon per source

> **Warning**
>
> If a URL requires login or credentials change, syncing may fail. Update access and retry.

## Group and manage sources

Group sources by product line, team, region, or document type. Sort by **newest**, **oldest**, **type**, or **name**. Each source offers:

* **Sync**
* **Rename**
* **Move to group**
* **Remove**

## Why isn't my agent using the sources I connected?

Several factors affect retrieval:

### Data structure

Sources splits content into 2000-character chunks with 500-character overlap. Very large documents or widely separated related sections may struggle more with relevance.

**What to do:**

* Restructure documents into smaller, tighter pieces.
* Repeat key headings or terms.
* Or curate the material as a managed topic for guaranteed usage.

### Update state

Two updates must be current:

* **Source Update** → keeps the data in each source fresh
* **Agent Update** → applies knowledge connection changes to the agent

Both can be triggered manually. Agent updates also run automatically every few minutes.

### Environments, variants, saved changes

Each source must be enabled in the correct **environment** and **variant**. Any edits must be **saved** before leaving the page.

## Conflicting information?

If the [FAQs](/agent-studio/knowledge/fa-qs/fa-qs) and Sources contain conflicting data, **the FAQs tab wins**. Content from the FAQs tab is always prioritized.

## Viewing Connected Knowledge in Conversation Review

When your agent retrieves content from Connected Knowledge during a conversation, you can see exactly which sources were used in [Conversation Review](/agent-studio/conversations/review).

1. Open a conversation in **Analytics > Conversations > Voice**.
2. In the **Diagnosis** dropdown, toggle **Sources** on.
3. Each turn where Connected Knowledge was retrieved shows a **Sources** tag beneath the agent's response, alongside any matched FAQs.
4. Click a source name to open an inline preview panel showing the exact text chunks the agent used.
5. Use **Open in Knowledge** in the panel to navigate directly to the source in the Knowledge area.

![sources-conversation-review](/_fern-files/_img/0b2adb6df3cb2e04637439aa6eda9419925b04544581f81347351ed12807d3a5.webp)

This is useful for:

* Verifying the agent retrieved the correct content for a given question
* Debugging cases where the agent's response seems inaccurate or incomplete
* Confirming that newly added or updated sources are being picked up

Combine the **Sources** and **Topic citations** diagnosis layers to see both Connected Knowledge and FAQs side by side for each turn.

## Behavior and configuration notes

* **Use PolyAI's Raven LLM** for best results – it paraphrases structured and unstructured content more naturally.
* Sources results are given ranking priority to ensure they surface alongside FAQs.
* Sources and FAQs data are merged at runtime.
  * Any [system-prompt style](/agent-studio/flows/overview) guidance applies to both.

## Related pages

#### [FAQs](/agent-studio/knowledge/fa-qs/fa-qs)

Create curated topics alongside connected sources. FAQs always take priority.

#### [RAG overview](/agent-studio/knowledge/fa-qs/rag)

Understand how retrieval-augmented generation works across your knowledge.

#### [Conversation diagnosis](/agent-studio/conversations/diagnosis)

Verify which knowledge sources were retrieved on each turn.