Breadcrumbs

Module settings

Once your product feed is connected and synced, the next step is to configure the Product Advisor Agents.
Each agent specializes in a different part of the conversation — from identifying the customer’s intent to recommending, comparing, or adding products to the cart.

These agents work together to create a seamless shopping experience powered by AI.


Most Agents should work out-of-the box even without any prompt. Use it only as a fine tuning to match your business logic after testing it out without it first.

Agents Prompt Settings

Overview of the available agents and their typical configuration options.

1️⃣ Triage Agent

Purpose:

Acts as the “traffic controller” for user queries.
It analyzes the customer’s message and decides which agent should handle the next step.

Routing rules examples:

  • General category questions and product selection help → Advisor Agent

  • Search for specific products with ≥ 3 attributes → Search Agent

  • Product information/specification questions → QnA Agent

  • Comparing products → Compare Agent

  • Add item(s) to cart → Add‑to‑Cart Agent

  • Off‑topic questions → Offtopic Agent

Prompt examples (Triage Agent Instructions):
Define custom routing rules logic:

“If the user asks about phone cases, skip Advisor and go directly to Search.”

“Never use the Search module.”

“Do not use the Advisor; use only the Search module.”


2️⃣ Advisor Agent

Purpose:

Guides users by asking a few questions (seek up to 3 key parameters) to refine product selection and find specific products.

  • Analyze categories and their parameters.

  • Leverage Buying History for personalization.

Workflow logic:

  1. Identify categories

  2. Load parameters

  3. Select key parameters

  4. Ask questions

  5. Build search query

  6. Interpret results

  7. Hand off to specialized agent (Compare, Add‑to‑Cart, QnA, Offtopic)

  • Output is specific to used channel:

    • Chat → Render a product carousel

    • Voice → Read out products and their key attributes)

Prompt examples (Advisor Agent Instructions):

“Always ask for shoe size in Sports -> Shoes category.”

“If the user asks about a TV, always confirm the screen size and brand.”

“Ignore product attributes x and y.”


3️⃣ Search Agent

Purpose:

When the user already provides enough information (in first message or after interacting with other agents), this agent directly finds products -> no clarifying questions.

  • Analyze the inputs to detect the most relevant categories.

  • Map user requirements to available product parameters.

  • Construct the search query.

  • Use Buying History to better interpret intent.

Workflow logic:

  1. Identify categories

  2. Load parameters

  3. Map request to parameters

  4. Build search query

  5. Hand off to the next agent

  • Output is a set of products summarized with regard to Instructions for Search Result Summarization. Out put is also specific to used channel:

    • Chat → Render a product carousel

    • Voice → Read out products and their key attributes)

Prompt examples (Search Agent Instructions):

“Ignore results below €5.”

“Always sort by most popular first.”

“For electronic devices, prioritize brand matching.”

“For shoe queries, always include Sport -> Shoes.”


4️⃣ QnA Agent

Purpose:

Handles detailed user questions about a specific product (based on available product information).

  • Identify the product from URL, name, or ID.

  • Load detailed product information from the database.

  • Provide precise answers based on product data.

  • Route complex questions to the appropriate agent.

Workflow:

  1. Identify the product in the query

  2. Fetch product info

  3. Find the answer in the data

  4. Respond to the customer

  5. Route follow‑ups as needed

Prompt examples (QnA Agent Instructions):

“Always include the product link with each response.”

“If the product’s data doesn’t contain the answer, apologize and offer to connect with support.”

“If asked about installation of product X, ask the user to email support@example.com and our technician will help.”

“If the customer uses word X, search for Z because X is slang not used in our catalog.”


5️⃣ Compare Agent

Purpose:

Helps users decide between two or more products. Tries to provide clear product comparisons with emphasis on key features, differences and similarities.

  • Return a comparison of at least two products, or escalate to an operator.

  • If there is no last search result available, request product URLs from the customer.

Prompt examples (Compare Agent Instructions):

“Highlight practical differences such as battery life or size.”

“For mattresses, prioritize comparing load capacity.”

“Ignore attributes that are identical across products.”

“If comparing products priced over €2,000, hand over to an operator.”

“Try to upsell the more expensive option.”


6️⃣ Add-to-Cart Agent

Purpose:

Adds selected items to the user’s shopping cart. Can confirm quantity or offer accessories.

  • A list of items to add to cart stored in $context as an array of objects: { item_id, amount }.

Prompt Examples (Add to Cart Agent Instructions):

“If the user adds more than 5 items, ask for confirmation.”

“Suggest compatible accessories when adding high-value items.”

“If cart value is more than 1000€, hand over to operator“


7️⃣ Offtopic Agent

Purpose: Handles unrelated questions politely and redirects to shopping context (or escalate to operator, or return to Triage agent).

Prompt examples (Offtopic Agent Instructions):

“If asked about weather, suggest suitable seasonal products.”

“If the question is unrelated to shopping, do not reply and offer assistance with products.”

“When facing complaints about competitors, politely highlight our advantages.”

“If asked about X, escalate to an operator.”


General Settings

General settings change behavior across all agents.

Index name

  • Selects which search index to use

Item types

  • Sets the type of your products. This setting affects how all agents behave
    and what terminology they use.

  • At the moment 'Products' for physical goods or 'Trips' for travel services are supported.

General instructions

  • High‑level guidelines for the whole system

  • Example: “You’re agent helping online store X, which sells running equipment. Your persona is man, communicate informally but politely.“

Buying history (Influences Advisor, Search, QnA agents)

  • Customer purchase history → Personalization input for better recommendations


💡Example End‑to‑End Flow

  1. Customer: “I need a lightweight laptop for photo editing under €1500.”

  2. Triage agent selects Advisor agent to continue: Advisor agent asks for screen size, RAM/SSD, preferred brand (max 3 key parameters).

  3. Advisor agent builds query and hand off to Search agent to retrieve candidates.

  4. Product Reply Agent summarizes results; chat shows carousel, voice reads highlights.

  5. Customer: “What is the difference between the first two?”

  6. Compare agent provides clear differences and practical trade‑offs.

  7. Customer: “Add the second one to my cart, 2 pcs.”

  8. Add‑to‑Cart agent stores [ { item_id: "...", amount: 2 } ]in context value, optionally proposes compatible accessories.