For the complete documentation index, see llms.txt. This page is also available as Markdown.

Setup your Sagent

Get your Sagent live with these essential steps.

Set up and launch Sagent

Sagent is an AI-powered guided selling chatbot that uses your product catalog to help shoppers find the right product through natural conversation. It draws on your product data and category knowledge to ask the right questions, understand what a shopper is looking for, and surface the most relevant recommendations.

This guide walks you through everything you need to get Sagent live — from creating the experience and connecting your product data, to configuring how it behaves, what it says, and how your customers will access it.

Before you begin

Make sure the following are in place before you start:


Set up

Create the Sagent and connect it to your product catalog.

1

Open the creation flow

Go to Experiences → Sagent → New Sagent.

2

Set the basic details

Set the Chatbot Name, Product Categories, if relevant, and the Chat Language.

3

Save the Sagent

Click Save.

4

Connect the data source

Open the Products tab, select the export feed Sagent should use, and click Sync.

Use Product Categories if your catalog contains distinct product groups — this helps Sagent stay focused on the right assortment rather than searching across unrelated products.

5

Test the data connection

Go to Setup → General, start a conversation in the preview panel, and confirm Sagent can access your product data before moving on.


Knowledge

Sagent's knowledge comes from the product data in your connected feed. The products that remain after your feed is synced are what Sagent draws on to answer questions and make recommendations.

1

Open the product catalog

Open the Products tab.

2

Review what Sagent can access

Use the search and filters to browse the catalog and confirm the right products are available to Sagent.

The quality of Sagent's recommendations depends directly on the quality of your product data. Well-structured properties and complete descriptions give Sagent more to work with.

3

Fix data issues at the source

Check for missing products, incorrect values, or incomplete descriptions. Review your feed setup if something looks off, then update the source data and sync again if needed.

If something looks off, the issue is likely in your feed setup. Use the links in Before you begin to review your product data configuration.


Behaviour

Control how Sagent conducts the conversation and when it shows recommendations.

1

Open the behavior settings

Go to Setup → Behavior.

2

Set recommendation thresholds

Set the Recommendation Thresholds to control when Sagent asks more questions versus showing results:

  • Maximum Matches — ask again if too many products still match

  • Maximum Questions — stop after a set number of questions

  • Minimum Details — require enough user input before recommending

3

Adjust tone and answer length

Adjust Tone of voice and answer length to match your brand.

4

Add guidance prompts

Add Guidance Prompts for any conversational behavior that standard settings don't cover.

5

Open the question flow

Open the Guidance tab.

6

Review suggested questions

Review the questions Sagent's algorithm has suggested, shown in priority order.

7

Reorder the questions

Drag and drop questions into the order you want Sagent to ask them.

8

Hide questions you don't want to ask

Use the eye icon to hide any questions Sagent shouldn't ask. Hidden questions can be restored at any time from the Questions hidden by you section at the bottom of the page.

9

Enable multi-select where needed

Enable Multi Select on any question where users should be able to choose more than one answer.

Use the answer chips to simulate user selections and preview how Sagent responds. This is for testing only — it doesn't affect your configuration.


Content

Update the copy and visual design so Sagent feels like a natural part of your experience.

1

Open the content settings

Go to Setup → Content.

2

Review the prefilled copy

Review the prefilled copy across the experience.

3

Update the copy

Update any text to match your brand.

4

Complete the privacy policy content

Complete the Privacy Policy Content field before launch.

5

Save your content changes

Click Save.

6

Open the design settings

Go to Setup → Design.

7

Set theme colors

Under Theme & Colors, set your Primary Color, Primary Text Color, Surface Color, and Secondary Text Color.

8

Set the gradients

Set the Aurora and Animation Shine gradients — these control the animated gradient on the loading screen and the shimmer effect that plays as each question appears.

9

Update the appearance

Under Appearance, upload your Top Bar Icon and set a Chat Background image or color.

10

Adjust the roundness

Adjust Button and Sagent Roundness to match your brand's style. The default is 6px across all elements.

11

Apply advanced styling if needed

Under Advanced, use the pre-filled CSS selectors to apply any further custom styling.

For a deeper look at every design option, see Configuring the Design tab.


Translations

If your Sagent needs to support multiple languages, set up your translations before launch — this covers the static copy users see throughout the experience.

1

Open the translations tab

Go to the Translations tab.

2

Select a language

Select a language from the dropdown. The languages available reflect what's set up in your Crobox account.

3

Review the fields

Review each field — the original English copy is shown alongside for reference.

4

Add or generate translations

Update the translations as needed, or click Generate Translations to auto-fill using AI.

5

Save the translations

Click Save.

Translations cover static interface copy such as the chat title, loading messages, and privacy policy text. Product content is drawn directly from your product feed and handled separately.


Deploy

A deploy is the entry point through which users open Sagent — whether that's a floating chat button on your website or a QR code in a physical store. Sagent comes with a default chat widget deploy already in place, so you're not starting from scratch.

1

Review the default deploy

Go to the Deploy tab. You'll see a Default website entry card already created — a floating chat button that appears bottom-right on the page. Click the card to review its settings and adjust if needed.

You need at least one enabled deploy before Sagent can be published.

2

Create a new deploy

Click + New Deploy.

3

Choose the deploy type

Set the Deploy Type — for example, a chat widget or a QR code.

4

Choose the Sagent flow

Set the Sagent Flow to control how the conversation starts:

  • Product discovery — starts the Guided Flow, with the full catalog in scope

  • Ask about a product — starts the Product Flow, with a specific product already in context

5

Name and configure the deploy

Give the deploy a name and description so it's easy to identify later. Configure any type-specific settings, such as page position for a chat widget or generating and downloading the QR code image.

6

Save the deploy

Click Save.

7

Manage existing deploys

  • Enable or disable a deploy using the toggle on its card. Disabling takes it offline immediately — useful if you need to pull it down temporarily without removing it entirely.

  • Duplicate a deploy as a starting point for a similar one — for example, a second QR code for a different product or location.

  • Remove a deploy to delete it permanently.

For a full overview of deploy types and configuration options, see Configuring deploys.


Validate before launch

Before going live, confirm that Sagent works correctly across the scenarios your users are likely to encounter.

1

Check catalog coverage

In the Products tab, use the search and filters to verify the right products appear for different property combinations.

2

Test Sagent in preview

In the Setup tab, chat in preview mode to test how Sagent responds to real queries.

3

Test the Product Flow

Open the chat with a specific product in context, as a user would from a product detail page or QR code.

4

Test the Guided Flow

Start from scratch with the full catalog in scope.

5

Test simple and complex queries

Test straightforward queries as well as ambiguous or complex ones.

6

Test additional capabilities

If you've configured additional capabilities such as the Comparison tool or Review summarizer, confirm these work as expected.

Why this matters: Even a well-configured Sagent can misinterpret answers or surface the wrong products if the data or question logic has gaps. Testing both simple and complex scenarios is the best way to catch those issues before your users do.


What's next

Once Sagent is live, you'll want to keep an eye on how it's performing:


Frequently asked questions

What makes Sagent's recommendations accurate?

Sagent draws on the product data in your connected feed and its built-in knowledge of your product category. The more complete and well-structured your product properties are, the better Sagent can match shoppers to the right products.

How do I know which export feed to use?

Use the feed that contains the product assortment you want Sagent to recommend from. If you're unsure which feed is correct, check with the person who manages your product data in Crobox.

What happens if I skip the Privacy Policy Content field?

You won't be able to take Sagent live. This field is required before launch, so it's worth filling it in during the Content step rather than at the last minute.

Can I change the data source after Sagent is live?

Yes. Go to the Products tab, select a different export feed, and click Sync. Be aware that switching feeds may affect the products Sagent recommends, so test after making the change.

What's the difference between the Product Flow and Guided Flow?

The Product Flow starts with a specific product already in context — for example, when a user opens Sagent from a product detail page or a QR code. The Guided Flow starts fresh, with the full catalog in scope. Both should be tested before launch to make sure Sagent handles each entry point correctly.

How many questions should Sagent ask before recommending products?

This depends on your catalog and your users. Use the Maximum Questions threshold in the Behavior settings to set a ceiling, and the Minimum Details threshold to ensure Sagent collects enough input first. A good starting point is to test with three to five questions and adjust based on how well the recommendations match what users are looking for.

Last updated

Was this helpful?