Training Hugo on your own resources will allow you drive your resolution rates up. Pertinent and complete resources demultiply the AI Agent's capacity to understand, reason and provide relevant answers to the many topics your customers may inquire about.

Hugo is a next-generation AI support agent, fully integrated with Crisp.

While conversational at its core, Hugo goes beyond answering questions. He understands customer intent and context to proactively guide users toward resolution — just like a human support agent.

Built to autonomously handle a large share of incoming conversations, Hugo can also perform real actions through your integrations, such as checking account details, tracking orders, or issuing refunds, always within the rules you define.

By engaging proactively and providing concrete guidance, Hugo helps resolve issues faster and frees up your team to focus on complex cases.

Haven't met Hugo yet? Check out official resources

Why training Hugo matters

There are different layers of configuration which are essential to improve and customize Hugo: defining a business prompt for Hugo to have context on your business, providing him with instructions to tweak its behaviour, creating routing rules for Hugo to perform tasks in certain situations, connecting integrations to an external system...

These all define Hugo's behaviour and scope of action, but they are only one face of the coin.

The second one is Hugo's knowledge. Data sources are what makes Hugo shine and allows him to handle conversations from start to finish and provide a great customer experience. Knowledge is power, and training Hugo on plentiful and resourceful data will turn him into an indispensible asset for your team.

You will find different types of training resources available, which all work together to bring cohesive information to Hugo:

• Web Pages → Allows you to crawl your own website and extract its content. Hugo is able to ingest it and search through it whenever it needs information.
• KB Articles → One of the many reasons to build a knowledge base with Crisp. Hugo is automatically trained on your articles whenever you create or update them.
• Files → Quickly import additional training material from a local files. Hugo supports TXT, CSV & PDF formats.
• Q&A Snippets → Build your own private FAQ for Hugo by creating short commonly asked questions/answers for Hugo to learn from. Quick and easy.

All of these sources allow Hugo to have proper information and knowledge around the questions commonly asked by your customers, making him knowledgeable about different topics also allows him to follow-up and engage more proactively with your customers, and help them beyond providing a simple response.

Training Resources

To start training Hugo, head over to Crisp, then navigate to ⚡️ AI Agent → Train, and follow this guide along to help you make the best out of them 🚀

Whenever Hugo converses with your users, he is able to search and use the resources ingested to extract the most pertinent information. Having resources on all of the most common topics brought up by your users is the guarranted way to help Hugo provide factual and reliable responses, at all times.

Web Pages

You likely already have a website full of quality content about your product or business, just ready to be ingested by Hugo.

Wether you have a generic website, a custom-built documentation, a blog, a developer hub or any type of hosted content, you can import its content into Hugo by crawling it.

To get started, head over to Hugo, then to Train → Web Pages.

From there, you can crawl new domains but also manage your currently crawled ones. If you have a knowledge base built with Crisp, it will also appear there by default.

How to import your web pages content

Fairly simple! At the top of the page, simply locale the Add a new website section, and specify the domain of your website (e.g. acme.com)

Optionally, before starting the import, you can also apply a filter if you wish to restrict which pages of your website will be crawled. This is handy if you wish to avoid training Hugo on certain pages and epurate your training data to preserve exclusively the most valuable ones. Don't worry though, you are still able to disable certain page even after the crawling occured

To apply a filter, hit the Configure page filters button, then Add Rule to get started. Rules offer the following options:

• Path Prefix → To capture pages starting with a certain path, e.g. /docs/en/ would match pages such as https://crisp.chat/docs/en/article/getting-started
• Path Equals → To capture an exact page, e.g. the pattern /integrations/whatsapp would match this page https://crisp.chat/integrations/whatsapp
• Path Contains → To capture pages whose path contain a certain partern, e.g. how-to would match pages such as https://crisp.chat/how-to-setup
• Page Language → To capture all pages whose content match a certain locale (using the html lang atribute of the page
• All Pages → Gotta catch 'em all!

Here is a usage example to capture all pages starting with a certain prefix. This requires 2 rules:

1. Our first rule containing our prefix condition with our pattern
2. A second rule to ignore all other pages (pages not matching the first rule)

Knowledge Base

Hugo is also trained on your knowledge base articles. This is a just another great reason to host your helpdesk with Crisp, or to import your existing one from another provider.

Articles are the perfect type of resources for AI Agents. Often detailled, organized and with complete information.

How to import your knowledge base content

Your Crisp knowledge base is automatically syncrhonized with Hugo anytime you:

• create a new article
• update an existing article
• delete or unpublish articles

No extra action required on our end, simply create and publish your articles, and Hugo will ingest them in realtime.

You can view it from Hugo in Train → Web content, just like your own web content.

Files

Hugo can also be trained on TXT, CSV and PDF files. This is a quick and easy way to import extra content, such as manuals, snippets, documentation, etc.

To get started, head over to Hugo, then to Train → Files.

How to import your files content

To import files, simply drag and drop a file (or hit the dropzone to select it from your computer) and validate the import.

Different types of files have different usages and requirements:

• TXT → To dump text-based information into Hugo
• PDF → To import documents/manuals/presentations...
• CSV → To import Questions/Answer in bulk

Questions & Answers

The most efficient way to train Hugo is by writting Question & Responses. This essentially allows you to build an internal and private FAQ dedicated to Hugo.

This data source not only helps you provide accurate and detailled information to Hugo, but also to bring corrections whenever you observe room for improvements when he replies to certain types of questions.

To create them, head over to Hugo, then to Train → Questions & Answers.

How to import your questions & answers snippets

When creating snippets, you should phrase them as if you were building a FAQ:

• Question → As a user would commonly ask it, e.g. "Do you offer free shipping?"
• Answer → The ideal and informative example answer to that question, e.g. "We offer free shipping exclusively for orders above $50. Below that, shipping costs are evaluated during check-out."

For optimal results, your questions should reflect the common terminology and phrasing used by your customers.

Your answers should also be phrased in a clear way. Here are a few recommandations for you:

• Be straight and concise in your answers, avoid unnecessary information
• Avoid ambiguous phrasing as well as mixing information that could be confusing without proper context
• Focus on answering the question, your answer should not contain behavioural prompt or address Hugo itself.

Customers may ask the same question in different ways. It is therefore important that your snippet's question be as generic and simple as possible.

Here are some examples:

✘ Bad question:

I paid with my credit card, got debited and I want a refund please.

✔ Good question:

How can I get a refund?

Similarly, your answers should be clear, organised, and provide detailled information concisely addressing the question.

Here are some examples:

✘ Bad response:

Tell the user he can do a refund from their account dashboard if the order is recent

✔ Good response:

You can perform refunds from your account dashboard inside of: Account > Orders > View Details.

Note that to be eligible for a refund, a few conditions must be met:
- The refund must be requested less than 30 days since you've received it
- The items must still be in prestine condition

Frequently Asked Questions

Still have some questions which were not covered above? Perhaps you'll find our answer in this FAQ. Also checkout this article and master your workflows craftsmanship!

If your question does not figure here, share it with us directly by chat, we'll be available to assist you

[Web Pages] Can Crisp crawl websites with dynamically-injected content?

Our crawlers require that the content of your website be available statically.

When crawling your website, Crisp will not execute javascript, hence dynamic content will not be fetched. This can prevent us from detecting any content, but also discovering other links, or following client-side redirections.

If your website is using CSR, we strongly recommend switching to SSR or SSG to avoid any issues.

[Web Pages] What if my website has a firewall blocking crawlers?

We may fail to crawl your website if it is protected behind a firewall or of it doesn't allow crawlers (e.g. some Cloudflare advanced protection settings or Distil Networks).

In such cases, some pages (or your whole domain) may not be imported.

If you are using a Firewall and observing import issues, we recommend monitoring your server logs and whitelisting our crawler user agent.

[Web Pages] How many domains can I import?

You can import up to 10 domains per workspace.

Note however that you can only import/refresh them one at a time.

If you want to add multiple domains, please first wait for the previous domain to be done importing.

[Web Pages] Are files hosted on my website automatically imported during crawling?

TXT, CSV, PDF and other types of pages are not imported automatically during web pages crawling.

To import those files, use the "Files" import option instead.

[Web Pages] Does Crisp automatically refresh my web content training when I update my website?

We do not automatically track and update your web content automatically.

If your website has been updated and you wish to update Hugo's web pages training, you can manually refresh it by hitting the 🔄 button.

[Knowledge Base] Does Crisp support custom-built and external knowledge base?

If you wish to import your own self-hosted or custom documentation, you can crawl it in the Web Pages section.

You can also migrate your knowledge base to Crisp if it was built with one of the supported providers listed in this article.

[Files] How many files can I import?

You can import up to 100 files per workspace.

[Q&A Snippets] Should I always translate my Questions/Answers snippets?

You generally do not need to translate your snippets into all languages.

Hugo is natively able to understand and translate information from one language to another. While having sources in multiple locales can help him better assess the context and information provided, this is often marginal.

Having a snippet in different locales is however very useful if the information differ based on certain languages. For instance if you communicate different informations, instructions, or reference menu/links for specific locales.

[Q&A Snippets] Are my Q&A snippets visible to users?

Answer Snippets are not visible to your users.

They are kept private, and only visible to you and your team. They are solely used by Hugo to generate response messages, which he will rephrase and use depending on the context.

[General] Is Hugo trained on other customer's conversations?

No, Hugo is not trained on other conversations whatsoever when replying to your customers.

Your customers' conversations remain private at all time, Hugo only has access to the training resources mentioned in this article, as well as the ongoing conversation context. That is all.

On the other hand, the Copilot (the agent-facing AI) does have access to other conversations knowledge when your agents ask him questions.

The Copilot never replies to your customers though, it is here to assist your own agents exclusively.

Updated on: 01/02/2026