Training your agent (links, files, and text)

Training your agent is what makes HaloDesk useful.

Your agent can only answer questions based on the content you give it, so setting up the knowledge base correctly is critical for good AI responses and fewer escalations.


What “training” means in HaloDesk

Training an agent means:

  • Collecting content (links, files, text)

  • Storing it as the agent’s knowledge base

  • Teaching the AI to answer questions using that content

Each agent has its own, separate knowledge base.


Training with website links

When you create an agent using a website URL, HaloDesk automatically:

  • Finds all pages under that domain

  • Lists them as sources

  • Lets you choose which pages to include or exclude

You can:

  • Keep all pages selected

  • Unselect pages you don’t want the AI to learn from (for example: legal pages or marketing-only content)


Crawling single pages vs full domains

HaloDesk supports two link types:

  • Single page crawl
    Best when you only want the AI to learn from one specific page.

  • Domain crawl
    Best for help centers or documentation sites (for example: help.yoursite.com).

When crawling a domain, you can:

  • Set a page limit (for example: crawl up to 25, 50, or 100 pages)

  • Include only specific paths

  • Exclude specific paths

This gives you full control over what content is used.


Uploading files (PDFs)

You can upload PDF files as knowledge sources.

Common use cases:

  • Product documentation

  • Onboarding guides

  • Internal manuals

  • FAQs

Uploaded files are indexed and treated the same way as website content.


Adding custom text snippets

Text snippets let you manually add information that:

  • Doesn’t exist on your website

  • Should override or clarify existing content

  • Is internal or operational

This is useful for things like:

  • Support policies

  • Edge cases

  • Temporary instructions


Storage limits (important)

Each agent has a storage limit, and your organization has a total storage limit across all agents.

Storage is calculated based on:

  • Number of crawled pages

  • File sizes

  • Text content

If you hit a limit, you’ll need to remove sources or purchase additional storage.


Updating sources and retraining

You can edit your knowledge base at any time:

  • Add new links

  • Remove outdated pages

  • Replace files

  • Update text snippets

After making changes, click Train to retrain the agent so the AI uses the latest content.


Training tips for best results

  • Focus on clear, factual content

  • Avoid duplicate or conflicting pages

  • Keep marketing copy out of support agents when possible

  • Retrain the agent whenever your documentation changes


Was this article helpful?
© 2026 HaloDesk