> ## Knowledge Base Index
> Fetch the complete knowledge base index at: https://help.chrisp.chat/sitemap.xml
> Use this file to discover available pages before exploring further.
> Pure-Markdown content can be obtained by appending a '.md' suffix to the content URLs listed in the sitemap (without the trailing slash).

# Getting started with Hugo AI Agent

*Set up Hugo AI Agent to answer customer questions using your support knowledge and the guidance you give it.*

It can help with recurring requests while passing cases that need a person to your team, and connected tools can add customer-specific information. This guide walks through a first setup, realistic tests, and reviewing the conversations after launch.

---

# ${color}[#1575dd](I. Prepare Hugo for your support questions)

Start with a small set of questions your team answers often, such as delivery options, account setup, or how to use a feature. Choose requests with a clear, current answer and decide which cases should reach a person. This gives you a useful first setup to test before adding more processes.

#### ${color}[#405e80](Configure the agent)

As a workspace owner, open **⚡️ AI Agent → Agent → Settings**. Add a name, avatar, and business description, then save the identity section. Leave incoming-conversation activation disabled while preparing your first setup.

Describe what your business sells, who it serves, and where it operates. Use your own facts; this fictional example shows the useful context:

> We are The Cookie Factory. We bake artisan cookies in small batches using quality ingredients.
>
> We sell classic and seasonal flavors, plus personalized cookies for events, gifts, and celebrations.
>
> Customers can order through our online store at **thecookiefactory.shop** or visit our store in **San Francisco**. We offer worldwide shipping and collection from the store.
>
> We focus on freshness, presentation, and small-batch quality rather than mass production.

Review and save the [model and answer settings](https://help.crisp.chat/en/article/how-to-choose-and-configure-hugos-ai-model-1m4mm57/) and the data sources Hugo is allowed to use.

![Hugo Agent Settings with a sample name, avatar, and business description completed](https://screenshots.invalid/hugo-agent-settings-with-a-sample-name-avatar-and-business-description-completed.png =800xauto)

#### ${color}[#405e80](Add support knowledge)

Add the sources Hugo needs to answer your customers’ questions. Use **AI Agent → Train** for Q&A, web pages, and files; manage Crisp articles in **Knowledge Base**.

__The following types of sources are available:__
* **Questions & Answers** for short, recurring questions
* **Web Pages** for your website or documentation
* **Files** for text, Markdown, CSV, or PDF resources
* **Crisp Knowledge Base** articles, which synchronize while published and not hidden; removing, hiding, or unpublishing them removes that Knowledge Base resource

For the Cookie Factory example, Hugo needs the delivery and collection policies to answer routine shopping questions. A short Q&A can fill a missing answer, while an existing help article can explain a complete procedure.

Use current, factual answers that are suitable for customer replies, and include their source types in **Agent → Settings → Data sources**. Put behavior rules in Instructions instead of Q&A snippets. Follow [the training guide](https://help.crisp.chat/en/article/how-to-train-hugo-ai-agent-on-your-data-ijkdrl/) for import steps and limits.

#### ${color}[#405e80](Define behavior and handoff)

Create [Instructions](https://help.crisp.chat/en/article/how-to-write-hugo-instructions-for-ai-agent-and-copilot-1qc131v/) under **AI Agent → Guidance → Instructions** for your tone, boundaries, and response rules. Select the AI Agent scope for customer replies, then save and enable each instruction you want Hugo to use.

Use [Routing](https://help.crisp.chat/en/article/how-does-hugo-routing-work-47o06k/) when a situation should start a Workflow or move a conversation to another inbox. Review [escalation settings](https://help.crisp.chat/en/article/how-to-configure-escalation-with-hugo-ai-agent-sqbm34/) so customers can reach your team when Hugo cannot help.

__For example:__
* a delivery policy is **knowledge**
* *“keep the reply friendly and explain the next step”* is an **Instruction**
* *“send custom bulk-order enquiries to Sales”* belongs in a **Routing rule**

Keeping each in the right place makes it easier to update the setup later.

---

# ${color}[#1575dd](II. Test, activate, and review)

Set a clear first rollout boundary: the requests Hugo should handle, the channels it should answer, and the team that will take over when it cannot continue. Check answer quality and usage together before expanding that boundary, since a working answer still needs an available budget and a reliable handoff.

#### ${color}[#405e80](Check your usage budget)

Open **AI Agent → Agent → Billing** to review available credits and usage. Hugo uses credits for the AI work it performs, so costs vary with the model and the request. Playground tests also use credits.

On a paid workspace, an owner can select **Enable pay as you go**, enter an additional-usage limit, and click **Save** if Hugo should continue beyond the included credits. Check the displayed limit afterward. Pay as you go is unavailable during a trial.

Without available credits or additional usage, automatic AI answers stop and Hugo follows the configured exhaustion handoff. This normally sends the conversation toward your team; eligible workspaces can use a dedicated Workflow instead. Review [credit usage and spending controls](https://help.crisp.chat/en/article/how-does-hugo-ai-pricing-and-billing-works-h78hkv/) for the full billing behavior and available limits.

#### ${color}[#405e80](Test the setup)

|| Connected tools can perform real actions during a Playground test. Use test records and suitable permissions for integrations before trying requests that change data.

In **AI Agent → Evaluate → Playground**, try a common support question, an incomplete request, and a case that should reach your team. Decide what answer or handoff you expect for each.

For the cookie store, that might be a correct collection-policy answer, a follow-up question when the customer only says *“delivery problem”*, and a Sales handoff for a custom event order. Check that each reply helps the customer move forward and does not promise something outside your policy.

Follow [Investigate and improve Hugo’s answers](https://help.crisp.chat/en/article/how-to-monitor-and-debug-hugo-ai-agent-conversations-with-observability-1q79tp1/) to inspect the search and tool activity, correct the relevant source or rule, and repeat the request.

![Hugo Playground with a sample customer question, the agent answer, and the inspection area for the resources it used](https://screenshots.invalid/hugo-playground-with-a-sample-customer-question-the-agent-answer-and-the-inspection-area-for-the-resources-it-used.png =1000xauto)

#### ${color}[#405e80](Enable Hugo for customers)

Open **AI Agent → Agent → Activation** and enable Hugo on incoming conversations. Choose the first response that fits your support process.

__The default-action choices are:__
* **Answer directly with AI Agent:** Hugo answers the customer first.
* **Directly route to your Inbox:** your team receives new requests without a first AI reply. Hugo does not evaluate its intent Routing rules before this handoff.
* **Directly route to a Workflow:** choose an enabled Workflow for a fixed process, such as collecting information before an answer or handoff. Check its entry block so it starts on the event you expect.

For an AI Agent default, choose **Automated inbox** or **Main inbox**. For a Workflow default, choose **Automated inbox** or **Keep Crisp inbox routing (default)**. Save the activation section, then select the incoming channels and save that section too. Choosing **Automated inbox** keeps automated requests separate from your team’s regular queue. Test a new customer message on an enabled channel and confirm that the chosen first response and destination work together.

Follow [How to configure Hugo’s activation](https://help.crisp.chat/en/article/configure-when-hugo-starts-resumes-and-resolves-conversations-draft0011/) for office hours, returning customers, and the complete activation settings.

![Hugo Activation showing the default-action choices, selected channels, and Automated inbox destination for a first rollout](https://screenshots.invalid/hugo-activation-showing-the-default-action-choices-selected-channels-and-automated-inbox-destination-for-a-first-rollout.png =800xauto)

#### ${color}[#405e80](Review the first conversations)

Read a few real conversations after launch, including successful answers and handoffs. Check the customer’s original request as well as Hugo’s last reply.

__Look for these useful outcomes:__
* **A complete answer:** the customer can follow the advice and finish the task.
* **A useful follow-up question:** Hugo asks for information it needs instead of guessing.
* **An appropriate handoff:** the team receives a case that needs human help, with enough context to continue.

When Hugo records an escalation reason, read it in the Inbox private note or the Observability details. Playground tests show the recorded reason in their activity details. Some fallback handoffs have no written reason; inspect the preceding steps in those cases. Check it against the conversation before deciding whether the source, guidance, or handoff needs changing.

Use [Hugo Analytics](https://help.crisp.chat/en/article/how-to-use-hugo-analytics-to-improve-your-ai-agent-6ymdxs/) to find recurring outcomes, then investigate individual examples in Observability. Correct the relevant source or rule and repeat the question in the Playground before expanding your setup.

Add [Workflows](https://help.crisp.chat/en/article/getting-started-with-workflows-hrmysb/) when a request needs a fixed intake or sequence of actions. Add [integrations](https://help.crisp.chat/en/article/how-to-use-hugo-integrations-6z9yhh/) when the answer depends on current customer data. For example, a shipping policy belongs in your knowledge sources, while [an individual order’s status](https://help.crisp.chat/en/article/how-to-create-an-ai-wismo-chatbot-with-crisp-18grv8j/) comes from the connected store.