Getting started with Hugo (AI Agent)

...First time? This article will guide you step-by-step to get your AI Agent setup and deployed in just minutes.

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 this dedicated article.

Setting up Hugo

In order to get Hugo and up and running, just a few things are needed:

Configuring Hugo's identity and settings
Providing it with some guidance (our instructions and routing rules)
Importing some resources to train it for your use-case

Let's go through them step-by-step and share a few good practices to help you along the way.

To get started, head over to Crisp and hit ⚡️ AI Agent from the main menu on the left.

Step 1 - Configuration

Inside Hugo, navigate to Agent > Settings.

First off, we must configure the AI Agent's identity.

You can start by uploading an avatar (which users will see when interacting with it) and it's name.

This next step is crucial: we must specify a business description.

You'll want to provide an introduction to your business, just as if you were describing it online to a new audience unfamiliar with it yet.

✔ Example of an efficient business description:

We are « The Cookie Factory », an artisan consumer food brand specializing in freshly baked cookies made in small batches using high-quality ingredients.
Our brand emphasizes craftsmanship, freshness, and thoughtful presentation over mass production.

Products range includes classic staples and rotating seasonal recipes, with optional personalization for events, gifts, and celebrations.

We sell directly to consumers through our online store (thecookiefactory.shop) and via our physical flagship store in San Francisco.
We offer worldwide shipping as well as in-store pickup.

The business description allows the AI Agent to familiarize with your business and the services it provides. It plays a major part in the AI's ability to understand and infer proper context when interacting with your customers.

Select your preferred AI model among the available options

Hugo works with any of the models available (GPT, Claude, Gemini, Mistral...), pick your poison! Don't worry, you can change at any time, and you will be able to test them all out in the Playground later.

In that section, you can also decide if Hugo should be able to read and understand images sent by users, and transcribe audio messages.

Finish up the configuration in that menu based on your preference

Enable the data sources you wish Hugo to use
Enable the default escalation rules when the Hugo doesn't have an answer or thinks the user might be getting frustrated
Choose the delay after which Hugo should auto-resolve the conversation if the user didn't follow up to its response

Satisfied users commonly continue their journey without acknowledging that their issue was solved, hence auto-resolving them can be handy to keep your inbox tidy.

Step 2 - Providing Guidance to Hugo

Guidances are the instructions (behavioural prompts) we give Hugo to personalize the way it replies to users, and where we configure routing actions for certain situations.

Navigate to Guidance > Instructions to create your first instruction.

Instructions are where you prompt the behaviour Hugo should adopt when replying to your customers and handling certain situations, this is essentially the AI's playbook it will follow at all times.

To create one, simply name it and write down your prompt.

✔ Example of a clear instruction :

Never offer medical or health advice. If a user asks about allergies, intolerances, or health impacts of our cookies, do not speculate or give recommendations.
Instead, provide only factual ingredient or allergen information if explicitly available, and advise customers to consult a qualified health professional for medical or dietary advice.

Good instructional prompts should be imperative and avoid ambiguity. Be specific and provide the necessary details to Hugo for the best results

✘ Example of an inproper instruction :

We ship our cookies worldwide. Estimated delivery is between 2 and 4 days in the US, and 5 to 10 days for the rest of the world.

Instructions define how the AI thinks and behaves, they should not be used to dump knowledge or facts. These would instead be training resources (e.g. Question/Answer snippets)

We can now also create our first routing action in Guidance > Routing

Routing actions are used when you wish Hugo to perform a specific task when it encounters certain situations. It can for instance be used to automatically escalate conversations when users are inquiring about quality issues with your products, or start a workflow when they request a marketing partnership.

To create one, simply name your rule, describe the situation (a topic, an emotion...), and select the desired action you wish Hugo to execute.

✔ Example of a good routing prompt:

The customer asks about personalized cookies for events (e.g. weddings, birthdays, corporate gifts, bulk quantities).
Does not apply if the customer asks about the delivery status of their personalized cookies order.

We could then select the routing action Escalate > Inbox "Sales"

A good routing prompt describes a situation clearly and precisely. It also covers potential cases where you wish to make exceptions to avoid ambiguity.

✘ Example of a bad routing prompt:

The customer is inquiring about their order.

Your description shouldn't be too broad and unspecific. In this example, it isn't clear what we are referring to: Issues to place an order? refunds? delivery information? Always be specific.

Step 3 - Importing training resources

We now need to provide Hugo with data and resources to help him accurately answer your customer's questions.

In Hugo, head over to the Train section.

Question & Answers allow you to create short snippets

Snippets are an essentials source of information for your AI Agent, use them without moderation.

They should contain two information: a question (the title/context of your snippet) and its answer (similarly to a short article).

✔ Example of a good Q&A snippet:

Question: What are your opening hours?
Answer: We are open Monday to Saturday, from 9:00 to 18:00 local time. Our store is also closed during public holidays.

Q&A Snippets should be exclusively informative and contain a question (the title) and its answer (the content). They are essentially the private FAQ of Hugo, and as such should not contain behavioural prompts (instructions, tone...). This would instead be the role of "Instructions".

Import your website content

Hugo can crawl your website to extract and learn from its content. Simply specify your URL, start the import, and let Hugo ingest everything while you keep configuring everything else

You can also specify a filter when importing your website, this can be useful if you wish to ignore certain pages and restrict Hugo to only the content that truly matters for better relevancy.

You can also optionally import file resources

Hugo accepts 3 different formats: CSV, PDF and TXT files.

Hugo is also automatically trained on your Crisp knowledge base, whenever you create, delete or update an article.

Feel free to take the time to progressively build one, this is one of the most valuable source of information for your AI Agent.

Ready to get started?

We of course recommend that you keep on training your AI Agent with additional resources and explore the additional configuration settings.

Once this is done, and you feel ready to test and deploy Hugo, you can simply follow those final couple of steps.

Head over to Evaluate > Playground to test Hugo for yourself

You can easily create and reset new conversations to ask different questions as your users would, and understand the area of improvements required on your end (new training resources, additional instructions...)

You are also able to experiment with different AI models to see which one best fits your use-case.

Once ready, go to Agent > Activation to deploy Hugo live

Tick the Enable agent for new conversation option
Select your default target ("Answer with AI Agent" for Hugo to automatically handle conversations)
Optionally choose specific conversations channels Hugo should exclusively handle (everything else will automatically be attributed to human agents instead)

Once activated, new incoming conversations will automatically be routed to Hugo inside of the "Automated" inbox. If agent intervention is required, conversations will automatically be escalated to your main inbox (and accordingly with your sub-inbox routing rules)

Navigating the Hugo interface

Let's now draw a map of the Hugo interface to better understand how each part connect to the other, and get an overview of the different configurations possible.

Evaluate

Playground → Quickly test your AI Agent and easily compare models
Analyics → Allows you to get insights on all AI behaviours (conversations handled, escalated, workflows triggered, topics encountered...)

Agent

Activate → To activate Hugo and decide how it should handle incoming conversations
Settings → All settings related to Hugo, ranging from its identity, model, sources to use, escalation handling...

Guidance

Instructions → Instructions are the prompts you feed the AI with to fine-tune its behaviour (tone, context, expectations...)
Routing → Describe a situation or topic, and decide what Hugo should do when it encounters it (starting a workflow, escalating to a human...)

Train

Questions & Answers → Create Questions/Answer snippets which the AI will use as its personal FAQ when answering common requests
Web pages → Crawl and import the content of your different websites, which the AI will be able to use as a resource
Files → Import additional training content as either TXT, CSV or PDF

Automate

Topic Detection → Describes topics that you would like to track in the Analytics, to get more insights on what your customers inquire about
Workflows Builder → Create your own custom workflows which the AI can then trigger through your "Routing" rules
Integrations & MCP → MCPs are the "tools" of AIs. Hugo offers native integration, but also allows you to connect to your own internal systems

Now with all that in mind, you're ready to continue strengthening your Hugo implementation!

Next steps

If you expected things to end here, we may still have a few more surprises before letting you go...

Below you will find an introduction to some of the other features shipped with Hugo 🤖

Topics & Analytics

Hugo is able to evaluate the topic of any conversation initiated by your customers. This allows you to monitor the type of conversations handled in your inbox, how Hugo handled them, and easily get insights on the amount of conversations it deflected and resolved for you.

This is also a great way to spot topics generating high volumes of conversations or identify areas of improvements in your setup and training resources provided.

You can create topics by heading over to Automate > Topic detection

Simply name your topic concisely, and prompt Hugo with a description of the topic/situation it should detect. Whenever Hugo is handling a conversation, it will evaluate your topics to see if the user's request matches any of them.

✔ Example of a good topic description:

The customer expresses negative feedback or dissatisfaction about:
- Our cookies (e.g. cookies not matching quality expectations, or issues with taste, texture or freshness)
- Shipping and delivery issues (e.g. missing or damaged goods, late delivery, lost package...)
- Poor customer service experience (e.g. frustration about our refund policies or the help they received)

Similarly to the "Routing", topic descriptions should be detailled and precisely describe the situations you wish Hugo to detect.

You can then monitor those topics and the paths taken by Hugo inside of Evaluate > Analytics

View the topic reccurency of conversations received
Compare escalation and deflection rates
Analyzing the steps and routings taken by Hugo when handling those conversation

Build Workflows

Hugo is also able to start workflows (formerly known as "scenarios") to achieve various use-cases such as collecting information, booking meetings and demo requests, offering different choices to your users, and much (much) more.

There are a couple of ways to integrate workflows with Hugo:

Automatically routing all conversations through a workflow, inside of Agent > Activation by switching the default target to Directly route to a workflow
Describing the different situations for which Hugo should start a specific workflow by configuring routing rules inside of Guidance > Routing

Workflows can also hand the conversation back to Hugo once their flow has been completed.

You can find out more about Workflows and their many usages inside of this dedicated article.

Explore integrations & MCP

By now, you understand why Hugo is an AI Agent, not just a chatbot: he can analyze conversations, understand text, images, and audio, make decisions, and know when to reply, route, or escalate — just like a human agent.

But... Hugo can also use tools 🛠️

MCP servers are the toolbox of AI Agents. It allows them to perform tasks which until now were only possible by human agents:

Look up a customer account details (profile, subscriptions, status...)
Check order status in real time
Cancel or modify an order
Fetch product information and stocks accurately
Retrieve invoice details or perform refunds
...and much more by connecting to your database to perform custom actions

In a nutshell, MCP are to AI Agents what your back-office is to your human agents.

They allow you to automate and delegate some of the most repetitive, time-consuming support tasks, while keeping full control over what the AI can and cannot do.

Crisp exposes various native ones with our different integrations (Stripe, Shopify, Woocommerce, Status Page...) but also allows you to easily and securely build your own, and connect it to Hugo.

You can learn more about MCP and their setup in this dedicated article.

Updated on: 19/01/2026

Was this article helpful?

Thank you!