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

# Knowledge overview

> Teach your AI agent your business. Five source types, one source of truth.

**Knowledge** is the content your AI is allowed to answer from. Without knowledge, your agent is polite and generic; with good knowledge, it's specific, accurate, and cites sources.

## Five kinds of sources

<CardGroup cols={2}>
  <Card title="Website crawl" icon="globe" href="/knowledge/website-crawl">
    Paste a URL. Keloa crawls the site and indexes every page.
  </Card>

  <Card title="File upload" icon="file" href="/knowledge/file-upload">
    Drop in PDFs, DOCX, TXT, or Markdown.
  </Card>

  <Card title="Text snippet" icon="align-left" href="/knowledge/text-snippet">
    Paste plain text — a policy, a pricing sheet, a FAQ.
  </Card>

  <Card title="Q&A pairs" icon="comments" href="/knowledge/qa-pairs">
    Write canonical question / answer rows. The highest-precision format.
  </Card>

  <Card title="Shopify" icon="bag-shopping" href="/knowledge/shopify">
    Pages, products, articles synced from your store.
  </Card>
</CardGroup>

## How knowledge works

When a message arrives:

1. Keloa converts the incoming question into a semantic vector.
2. It searches your knowledge and pulls the top-matching passages.
3. Those passages + system prompt + conversation go to the AI.
4. The AI answers, citing the passages used.

You see those citations in the conversation's **Sources** panel.

## Scoping by agent

Every knowledge source belongs to a specific **AI agent**. So you can:

* keep product pages with the pre-sales agent
* keep policy PDFs with the returns agent
* keep an internal snippet on a dedicated "internal support" agent and never show it externally

Change the agent a source is attached to any time — it re-indexes in seconds.

## Statuses

A source moves through these states:

| Status      | Meaning                                                                 |
| ----------- | ----------------------------------------------------------------------- |
| **Queued**  | Created, waiting for worker                                             |
| **Syncing** | Being crawled or parsed right now                                       |
| **Synced**  | Indexed and live — the agent can use it                                 |
| **Failed**  | Something went wrong. Click **Retry** or open the source to see details |

## Your storage budget

Each plan includes a knowledge allowance measured in **characters** of extracted text — roughly 2,500 characters per page. You'll see a bar at the top of the **Knowledge** page showing usage vs cap.

| Plan     | Knowledge characters        | Pages / source       | Pages / month |
| -------- | --------------------------- | -------------------- | ------------- |
| Starter  | 100,000 (≈ 40 pages)        | 25 (single URL only) | 25            |
| Growth   | 1,000,000 (≈ 400 pages)     | 120                  | 120           |
| Business | 10,000,000 (≈ 4,000 pages)  | 1,200                | 1,200         |
| Scale    | 75,000,000 (≈ 30,000 pages) | Unlimited            | Unlimited     |

If you hit the cap, add a bigger plan or delete sources you no longer need.

## Source-type availability per plan

Not every source type unlocks at every tier:

| Type              | Starter                 | Growth                  | Business                 | Scale                    |
| ----------------- | ----------------------- | ----------------------- | ------------------------ | ------------------------ |
| **Website crawl** | 1 site, single URL only | Unlimited, full-site    | Unlimited, full-site     | Unlimited, full-site     |
| **Text snippet**  | 5 max                   | Unlimited               | Unlimited                | Unlimited                |
| **File upload**   | 1 file, ≤ 5 MB each     | Unlimited, ≤ 50 MB each | Unlimited, ≤ 100 MB each | Unlimited, ≤ 200 MB each |
| **Q\&A pairs**    | Not available           | Available               | Available                | Available                |
| **Shopify**       | Not available           | Available               | Available                | Available                |

Auto-recrawl scheduling: Starter is manual-only; Growth = monthly; Business = weekly; Scale = daily. See [Website crawl](/knowledge/website-crawl#keeping-content-fresh) for the cadence ladder.

## Quality beats quantity

Surprising truth: 20 well-written Q\&A pairs will outperform a 1,200-page crawl for the **top 80% of customer questions**. A focused, up-to-date knowledge base answers faster and more accurately than a huge messy one.

Our rule of thumb:

1. Start with a **website crawl** for broad coverage.
2. Add **Q\&A pairs** for the 20 most common questions.
3. Add **file uploads** for policies and specs that don't live on the public site.
4. Prune stale sources every quarter.

## Next

<CardGroup cols={2}>
  <Card title="Add a website crawl" icon="globe" href="/knowledge/website-crawl">
    The fastest broad coverage.
  </Card>

  <Card title="Write Q&A pairs" icon="comments" href="/knowledge/qa-pairs">
    Hand-tuned, highest precision.
  </Card>

  <Card title="Manage sources" icon="folder-open" href="/knowledge/managing-sources">
    Recrawl, retry, delete, reassign.
  </Card>

  <Card title="Test your agent" icon="flask" href="/agents/test-agent">
    Confirm new knowledge shifts replies the right way.
  </Card>
</CardGroup>