What is a Data Search node?

A Data Search node lets your AI agent answer questions using structured data from CSV or Excel files.

Use it when the information your AI needs is organized into rows and columns, such as:

• Product catalogs
• Pricing tables
• Inventory
• Customer records
• Orders
• Policies and claims
• Broker or account data
• Operational reports
• Other structured datasets

The node can work with a single file or with multiple related datasets that are connected by a shared key.

What visitors can ask

Visitors can ask plain-language questions about the data.

You do not need to write a separate prompt that describes every possible question.

The node can handle tasks such as:

Look up a row

Examples:

1What's Sarah's email?
2What is the price of SKU-4021?
3What is the status of order 12345?

Summarize a column

Examples:

1How many managers do we have?
2What is the average price?
3What is the total revenue?
4What is the highest or lowest value?
5Show me the top 5 by impressions.

The Data Search node can find matching rows or calculate common summaries such as:

• Count
• Sum
• Average
• Minimum
• Maximum
• Top N

Answers stay grounded in the selected data.

Data Search vs. Document Search

Use Data Search when your information is structured as rows and columns.

Use Document Search when the information lives in documents, PDFs, articles, manuals, or other long-form text.

A simple rule:

Rows and columns → Data Search

Pages of text → Document Search

Choose a data source mode

The Data source section has two modes:

• Single file
• Multi-dataset

Choose the mode based on how your data is organized.

Single file

Use Single file when all the information needed for the search is contained in one CSV or Excel file.

Select the file from the dropdown.

If the file has not been added yet, use Upload file to add it.

For example, one product file might contain:

• SKU
• Product name
• Category
• Price
• Availability

A visitor could then ask:

1What is the price of SKU-4021?
2Which products are currently available?
3What are the top 5 most expensive products?

Multi-dataset

Use Multi-dataset when the information is split across two or more structured files and those files share a common identifier.

This lets Data Search retrieve and analyze information across related datasets instead of requiring everything to live in one spreadsheet.

For example:

Dataset 1: Policies

Dataset 2: Claims

Both datasets contain the same underlying policy identifier.

That relationship lets the AI answer questions such as:

1What claims are connected to policy P-1001?
2What is the total claim amount for policies handled by Smith Agency?
3Which broker has the highest total claim amount?

How multi-dataset search works

Multi-dataset mode links files through a Join key.

A join key is a column in each dataset that represents the same identifier.

For example:

The column names do not have to be identical.

What matters is that the values represent the same thing.

For example:

1Policies.Policy Number = P-1001
2Claims.Policy ID = P-1001

AskHandle can use that shared value to connect the records.

Configure Multi-dataset mode

To configure multiple datasets:

1. Open the Data Search node.
2. Under Data source, select Multi-dataset.
3. Configure Dataset 1:
   • Enter a label
   • Select the file
   • Enter the join key
4. Configure Dataset 2:
   • Enter a label
   • Select the file
   • Enter the join key
5. Click + Add dataset if you need to connect additional files.
6. Save the node.

You can also use Upload a new file if one of the required datasets has not been added yet.

Dataset labels

Each dataset has a Label.

The label is for your own clarity and helps identify what the file represents.

Examples:

• Customers
• Orders
• Policies
• Claims
• Products
• Inventory
• Brokers

Choose labels that make the relationship between datasets easy to understand.

Choosing a join key

The join key is the most important part of a multi-dataset setup.

Use a field that:

• Exists in each related dataset
• Represents the same entity
• Contains matching values
• Is stable and consistent
• Is specific enough to connect the correct records

Good join keys include:

• Customer ID
• Policy number
• Order ID
• Product SKU
• Claim number
• Account number
• Email address, when it is unique and consistent

Avoid broad fields such as:

• First name
• City
• Category
• Status
• Date

Those values can repeat across many records and may not identify the correct relationship.

Cross-dataset retrieval and calculations

Multi-dataset mode is useful for more than simple lookups.

Once datasets are connected by a shared key, Data Search can retrieve related rows and calculate summaries across the connected data.

For example:

• Total claims by broker
• Average order value by customer segment
• Number of orders per account
• Highest sales value by product category
• Top policies by claim amount
• Sum, average, minimum, maximum, count, or Top N across connected records

This lets the AI answer questions that require information from more than one file.

Search synonyms

Use Search synonyms when visitors may use different terms for the same value.

For example:

1NYC = New York City
2US = United States
3SKU = Product Code

Click + Add group to create a synonym group.

This can improve matching when the wording in the visitor's question differs from the values used in the dataset.

Show Found Data to Users

Enable Show Found Data to Users when you want matching rows to be shown as a clickable attachment.

If this setting is off, the matching rows are not shown directly to the visitor.

Either way, the found data remains available as hidden context for the next node.

Use Data Search with AI Answer

A useful pattern is:

1Data Search → AI Answer

Data Search retrieves the matching data first.

The found rows then remain as hidden context for the next node.

An AI Answer node placed after Data Search can use that context to answer follow-up questions without running the same search again.

This is especially useful when the user wants to continue discussing the result.

For example:

1. The user asks for policy P-1001.
2. Data Search finds the policy and connected claim records.
3. AI Answer explains the result.
4. The user asks a follow-up question.
5. AI Answer can use the already-found data as context.

Allow users to upload a file in Chat Portal

You can enable Allow users to upload a file in Chat Portal.

When enabled, visitors using Chat Portal can attach a structured file during the conversation and Data Search can use that file.

This option is intended for Chat Portal.

It is not recommended for Web Messenger, API, or WhatsApp.

Answer tone

You can choose how Data Search answers should sound.

Available options include:

• Friendly
• Formal
• Brief
• Custom

This changes the style of the answer, not the underlying data retrieval.

Most users can leave this on Friendly unless the use case calls for a different tone.

Name on canvas

Use Name on canvas if you want to give the node a more specific internal name.

Examples:

• Product Lookup
• Policy Search
• Order Search
• Customer Data
• Claims Analytics

This helps your team understand what the node does when looking at the workflow.

Example: policy and claims search

Suppose you have:

Policies file

Claims file

Configure:

1Dataset 1 label: Policies
2Dataset 1 join key: PolicyNo
3
4Dataset 2 label: Claims
5Dataset 2 join key: Policy #

The user can then ask:

1What claims are connected to P-1001?
2What is the total claim amount by broker?
3Which class has the highest claim amount?

Data Search can connect the relevant rows through the shared policy number before producing the result.

Best practices

• Use CSV or Excel files with clear column headers.
• Keep IDs and shared keys consistent across files.
• Avoid unnecessary duplicate rows.
• Use stable IDs instead of names when possible.
• Add synonyms when your users commonly use alternate wording.
• Use Multi-dataset only when the files have a meaningful relationship.
• Make sure all datasets use compatible values for their join keys.
• Test both direct lookups and summary questions.
• Place AI Answer after Data Search when you want richer follow-up conversations.
• Use Show Found Data to Users only when exposing the matched rows improves the user experience.

Important to know

• Data Search is for structured CSV and Excel data.
• Single file mode searches one dataset.
• Multi-dataset mode connects two or more related datasets.
• Datasets are linked through shared join keys.
• Retrieval and summaries can work across connected datasets.
• Data Search can calculate count, sum, average, min, max, and Top N results.
• Found rows can remain as hidden context for the next node.
• AI Answer can use that context for follow-up questions.
• For long-form text knowledge, use Document Search instead.