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

# Creating a Knowledge Base

> Step-by-step guide to creating a Knowledge Base in Timely.ai, adding content sources, and tracking processing status.

A **Knowledge Base** organizes the content that the agent queries in real time during conversations. Each base is independent and can contain multiple sources: documents, Q\&As, website pages, and YouTube video transcripts. The same base can be connected to different agents, and an agent can use multiple bases simultaneously.

## Step 1 — Access the Knowledge Module

1. In the sidebar, click **Knowledge** (or **Knowledge Bases**).
2. The screen displays all existing bases in the workspace, organized by creation date.

## Step 2 — Define the Base

Click **+ New base** in the upper right corner. A creation form will appear with two fields:

* **Name** — internal identifier for the base. Choose something descriptive that indicates the purpose: "Product FAQ", "Return Policies", "2025 Service Catalog".
* **Workspace** — if the account has multiple workspaces, select which workspace this base belongs to. The base will only be visible to agents in that workspace.

## Step 3 — Add Items

With the base created, add your information sources. The base supports four content types that can coexist in the same base:

### Text and Documents

Upload PDF or text files directly through the interface.

* The processor automatically extracts text from the file
* Content is split into chunks of \~4,000 characters with an overlap of \~400 characters
* Each chunk is vectorized and stored for semantic search

### PDF

PDFs with structured content — manuals, policies, catalogs — are processed the same way as text documents.

* PDFs with scanned content (scanned images) are not supported — the extractor requires selectable text
* Maximum file size is 50 MB
* The status changes from **Processing** to **Indexed** when vectorization is complete

## Step 4 — Indexing and status

After adding a source, content goes through the asynchronous processing pipeline. Track the status on the base screen:

| Status         | Description                                                                                              |
| -------------- | -------------------------------------------------------------------------------------------------------- |
| **Processing** | Content is being extracted, split into chunks, and vectorized. The item is not yet available for search. |
| **Indexed**    | Processing complete. The agent can now retrieve this content in conversations.                           |
| **Error**      | Processing failed. Click the item to view the error details and try again.                               |

## Step 5 — Update or Manage

After the initial creation, the base can be maintained in three ways:

* **Add new sources** — click **+ Add** inside the base to include new documents, Q\&As, URLs, or videos without removing existing content.
* **Edit Q\&As** — question-and-answer pairs can be edited directly through the interface. Re-indexing occurs automatically after saving.
* **Remove outdated content** — delete documents or Q\&As that are no longer valid. Deleted content stops appearing in searches immediately.

## Tip

<Note>
  Start with manual Q\&As for the 10 to 20 most frequent topics. They are indexed instantly — no asynchronous pipeline — and have high precision because you control exactly what the agent will retrieve. Then supplement with documents for extensive, contextual content.
</Note>