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

# Keep knowledge current

> Re-read your site, exclude pages the agent shouldn't answer from, remove stale sources, and deal with stuck indexing.

Knowledge is a snapshot, not a live copy of your site. WireDesk doesn't re-read your website on a schedule. A page you changed this morning isn't changed in the agent until the site is read again. Old content can do as much harm as missing content: a promotion that ended in March is worse than no answer at all.

## When to update

* **After a meaningful change to your site**: new pricing, new hours, a new product line, a changed policy.
* **When [Test a question](/docs/knowledge/test-a-question) returns an old passage.**
* **When conversations show the agent quoting something that's no longer true.**

For a single changed fact, adding a [note](/docs/knowledge/notes) is quickest. For a lot of changed content, re-read the site.

## Re-read a website

There's no separate re-crawl button. You re-read a website in one of two ways:

* **Save a change to its exclusions.** Changing which pages are left out (below) and clicking **Save & re-read the site** reads the whole site again. The agent keeps answering from the current pages until the new read is finished.
* **Remove the source and add the address again.** This is a full fresh crawl. The agent has no knowledge from that site until the crawl finishes, which is usually a minute or two.

See [Re-crawling](/docs/knowledge/website#re-crawling) for the details.

## Exclude pages from a website

Plenty of what a crawl reaches is noise the agent should never answer from: sign-in pages, terms and conditions, blog archives, event listings, expired promotions. Leaving them out also gives the pages you do care about more room within the crawl's [limits](/docs/knowledge/website#limits).

<Steps>
  <Step title="Open the website source">
    On **Knowledge**, click a ready website source. You'll see its pages grouped by the first part of the path, such as `/blog` or `/services`, with each page's size.
  </Step>

  <Step title="Find the pages">
    Use **Search pages** to filter by path or title. The sort button switches between sorting by path and sorting by size, which makes one huge page easy to spot.
  </Step>

  <Step title="Untick what should be left out">
    Untick a single page, or untick a group to leave out everything under that path. Excluded paths are listed under **Left out of the knowledge base**.
  </Step>

  <Step title="Save">
    Click **Save & re-read the site**. It takes a minute. The agent answers from the current pages until the new read finishes. Once it has, the excluded pages are gone from the page list and from the agent's knowledge.
  </Step>
</Steps>

How exclusions match:

* **A path covers everything beneath it.** Excluding `/tools` also excludes `/tools/pricing-calculator`, but not `/toolsmith`.
* **Case, trailing slashes, query strings and `#` fragments are ignored.**
* **Exclusions stick.** They're saved on the source and applied every time the site is read.
* **Excluded pages are never fetched.** The page and text limits they would have used go to the rest of the site.
* **You can exclude the page you entered.** The crawler still reads it to find links to other pages, but doesn't index it.
* You can't exclude the whole site (`/`). To stop using a site, remove the source.
* Up to 200 excluded paths per source.

To put a page back, click it in the **Left out of the knowledge base** list, or tick it again in the page list, then save. You can't change exclusions while the site is being read. You'll see "this site is being read right now — try again when it finishes".

Exclusions only apply to websites. To stop the agent using a file or a note, remove it.

## Remove a source

Click the bin icon on the source's row. The dialog asks **Remove "…"?**. Choose **Remove** to confirm or **Keep it** to cancel. The source's passages come out of the knowledge base straight away, and the agent stops answering from them. There's no undo: to get it back, you add the website, file or text again.

## Stuck or interrupted indexing

Websites and files are indexed in the background. If a source still shows "crawling the site…" or "processing…" after 30 minutes, WireDesk treats the job as interrupted:

* **A website** may be tried again automatically, up to three attempts. After that it fails with "this site could not be indexed after 3 attempts — it may be too large or too slow to crawl". Excluding large sections you don't need usually helps.
* **A file or a note** can't be retried, because WireDesk doesn't keep the original. It fails with "indexing was interrupted — add this source again".
* You may also see "indexing was interrupted — remove this source and add it again".

In each case, remove the source and add it again. A failed re-read of a website that was already ready doesn't remove what the agent already knew from it.

## A regular routine

1. **Weekly:** read a sample of [conversations](/docs/workspace/conversations). Every "I don't have that to hand" is a question your knowledge doesn't answer. Write a note for each one.
2. **After a site change:** re-read the site, then check two or three affected questions in [Test a question](/docs/knowledge/test-a-question).
3. **Occasionally:** open each website source, sort by size, and exclude anything the agent doesn't need.

## Related

* [Train the agent on your website](/docs/knowledge/website)
* [Notes and text](/docs/knowledge/notes)
* [Fix a wrong answer](/docs/knowledge/fix-a-wrong-answer)
* [Conversations](/docs/workspace/conversations)