Retrieve exact answers from structured business data
Data Search lets an AskHandle agent search structured data stored in CSV and Excel files and return precise answers from the records inside them.
It is designed for information that lives in rows and columns rather than documents.
Do you have a new white BMW SUV under $70,000?
Matching rows| Stock ID | Make | Model | Year | Color | Mileage | Price |
|---|---|---|---|---|---|---|
| A102 | BMW | X5 | 2026 | White | 8 | 68,500 |
| B284 | BMW | X3 | 2025 | Black | 4,300 | 51,900 |
| C317 | Audi | Q5 | 2026 | White | 15 | 57,800 |
Customers can use it to look up individual records, filter matching rows, summarize columns, calculate common aggregates, or connect multiple datasets through a shared key and answer questions across them.
The user asks in plain language. Data Search finds the matching data and keeps the answer grounded in the uploaded files.
Data Search works with structured tabular data in:
The data should be organized into identifiable rows and columns.
A vehicle inventory file, for example, might contain:
A visitor does not need to know those column names.
They can ask:
Data Search can use the structured values to identify matching records and provide the relevant result.
Businesses often have important information stored in spreadsheets.
That data may represent:
Data Search makes that structured information available through conversation.
Instead of requiring someone to open a spreadsheet, find the correct sheet, filter the rows, and inspect the relevant columns, the user can simply ask the question.
For example:
What's the price of SKU-4021?
Do we have a white SUV under $45,000?
How many managers are in this file?
What is the average transaction value?
Show me the top five campaigns by impressions.
Data Search interprets the request, searches the connected data, and returns the relevant result.
Customers do not write a query language or create a prompt for each request.
The node is designed for ordinary conversational questions while keeping the answer tied to the data.
There are two major types of questions Data Search can handle.
Use Data Search to locate one or more records.
What's Sarah's email?
What is the price of SKU-4021?
Find apartments with two bedrooms under $4,000.
Show me vehicles with fewer than 20,000 miles.
Which locations are open on Saturdays?
The node identifies the rows that satisfy the request and uses them as the basis of the answer.
Data Search can also perform common calculations over the records in a dataset.
Supported operations include:
How many managers do we have?
What is the average order value?
What is the highest monthly revenue?
What is the lowest listed price?
Give me the top five campaigns by impressions.
These calculations are performed against the data in the file.
This allows users to ask analytical questions without manually calculating the result in a spreadsheet.
The purpose of Data Search is not to make users think in rows and columns.
The structure stays behind the conversation.
For example:
Visitor
Which two-bedroom apartments are available below $3,500?
Data Search
Searches fields such as:
Answer
There are three matching apartments. The lowest-priced option is Unit 12B at $3,150 per month.
The user asks naturally while the node handles the underlying structured lookup.
Some questions cannot be answered from one file.
The necessary information may be distributed across several datasets.
Data Search supports Multi-dataset search for these situations.
Customers can connect two or more CSV or Excel files through a shared key column.
That common value allows AskHandle to associate records across the datasets and answer questions that depend on information stored in more than one file.
Imagine an insurance operation with two datasets.
| Policy ID | Customer | Product | Premium |
|---|---|---|---|
| P-1042 | Acme Corp | Commercial | 8,250 |
| P-1187 | Northstar LLC | Property | 5,600 |
| Claim ID | Policy ID | Claim Amount | Status |
|---|---|---|---|
| C-901 | P-1042 | 2,100 | Closed |
| C-932 | P-1042 | 1,850 | Open |
| C-940 | P-1187 | 900 | Closed |
Both files contain:
Policy ID
That column becomes the join key.
The user can now ask:
How much has Acme Corp paid in premiums and what is the total value of its claims?
The customer name and premium may live in the Policies dataset.
The claim values may live in the Claims dataset.
Data Search connects the records through Policy ID and can use information from both datasets to produce the answer.
Each dataset in a multi-dataset configuration receives a join key.
The join key is the shared value that identifies how records in different files relate to one another.
Common join keys might include:
The exact field depends on the business.
The important requirement is that the datasets share a value that can reliably connect the corresponding records.
Without multi-dataset search, a professional may need to:
With Data Search, that data relationship is configured once.
The user can then ask the business question directly.
This is particularly valuable for teams whose operational data is already structured but fragmented across several exports or systems.
Products.csv
Shared key:
SKU
Inventory.csv
Question:
Which running shoes under $150 are currently in stock in Manhattan?
The product information comes from one file.
The stock information comes from another.
Properties.xlsx
Shared key:
Property ID
Units.xlsx
Question:
Which available two-bedroom units are in Brooklyn and under $4,500?
The property location is stored in one dataset.
Unit availability and rent are stored in another.
Customers.csv
Shared key:
Customer ID
Revenue.csv
Question:
What was total revenue from enterprise customers managed by Sarah last quarter?
The classification and account manager live in one dataset.
Revenue lives in another.
Data Search connects them through the customer ID.
Structured datasets can be precise, but customers do not always use the same wording that appears in a spreadsheet.
Data Search includes optional search synonyms to solve this.
Synonyms map alternate terms to the values used in your data.
For example:
NYC New York New York City
can be treated as equivalent terms.
Or an automotive dataset may contain:
Sport Utility Vehicle
while customers usually ask for:
SUV
A synonym group can associate those terms.
Other examples include:
Product terminology
TV ↔ Television
Property terminology
2 bed ↔ 2 bedroom ↔ two-bedroom
Location terminology
LA ↔ Los Angeles
Business terminology
Enterprise ↔ Large business
Synonyms are particularly useful when the values in the dataset use formal terminology, abbreviations, codes, or naming conventions that differ from everyday language.
Semantic understanding is useful for interpreting a question, but structured lookup often depends on matching the actual values in the data.
A dataset might contain:
New York City
while a customer enters:
Show me properties in NYC.
A configured synonym makes that relationship explicit.
This helps Data Search preserve the precision of structured lookup without forcing users to know the exact wording in the file.
The goal is simple:
Customers speak naturally. The data can remain structured.
When Data Search finds matching rows, those records can stay available as hidden context for the next node.
This makes Data Search useful as both an answering node and a retrieval step inside a larger agent.
For example:
User
Which SUVs do you have below $50,000?
Data Search
Finds the matching inventory records.
AI Answer
Uses those records to continue the conversation naturally.
The customer might then ask:
Which of those has the most cargo space?
The AI Answer node can use the records already retrieved as context rather than requiring the user to start over.
This creates a useful pattern:
Data Search finds the facts. AI Answer continues the conversation.
Like Document Search, Data Search includes answer-tone controls.
Available options include:
I found 14 matching vehicles. The lowest-priced one is a 2025 X3 at $41,900.
Fourteen vehicles meet the selected criteria. The lowest listed price is $41,900 for a 2025 X3.
14 matches. Lowest price: 2025 X3 at $41,900.
A customer can provide its own style instruction, such as:
Lead with the result. Use short sentences. Include prices with currency formatting and list no more than five records unless the user asks for more.
The data remains the source of truth while the customer controls the presentation.
The tone changes how the result is communicated.
It does not change the data used to produce the answer.
For example, the same result could be presented differently.
Data Search can also allow users to upload a file through Chat Portal.
When enabled, a visitor can attach structured data during the conversation and ask questions about that file.
Examples might include:
The product interface notes that this mode is intended for Chat Portal and is not recommended with the widget, API, or WhatsApp.
Use Data Search when:
It is especially useful for businesses that already have valuable operational data but need a simpler way for customers or staff to access it.
Both nodes search customer-provided information, but they are designed for different types of knowledge.
Best for written, unstructured information.
Examples:
The answer comes from passages of text.
Best for structured records.
Examples:
The answer comes from fields, rows, filters, and aggregates.
A useful rule is:
If a person would normally read the file, use Document Search.
If a person would normally filter, sort, look up, join, or calculate the file, use Data Search.
AI Answer provides broad conversational intelligence, instructions, web access, and optional skills.
Data Search has a narrower purpose:
retrieve and analyze exact structured data from the files connected to the node.
For example:
What are the differences between leasing and financing?
may belong in AI Answer or Document Search.
But:
Which vehicles in our inventory are available for lease under $500 per month?
depends on the dealership's actual structured inventory and belongs in Data Search.
The two can work together.
Data Search can calculate results that come directly from the dataset, but it is not intended to replace the more configurable Calculations skill in AI Answer.
A useful distinction is:
Data Search
Calculate something from values already stored in the dataset.
Examples:
AI Answer with Calculations enabled
Collect inputs from the user and apply a business-defined formula.
Examples:
Data Search analyzes existing structured data.
The Calculations skill executes a formula configured by the customer.
Router can send data-dependent requests directly to the appropriate dataset.
For example:
Router
Vehicle search
→ Data Search: Inventory
Pricing and policies
→ Document Search
Book test drive
→ AI Answer with Scheduling enabled
Human assistance
→ Human Handoff
This keeps each node focused on the type of work it handles best.
Use Document Search when the answer comes from written documents.
Use AI Answer when the conversation requires broader reasoning, live web information, or action skills.
Use AI Answer with Calculations enabled when the business needs to collect inputs and run a custom formula.
Use AI Form when information should be collected through a controlled sequence of questions.
Use Router when different requests should reach different data sources or workflows.
Use Human Handoff when the conversation needs to move to a person.
Data Search turns structured business data into something users can interact with conversationally.
Upload a CSV or Excel file. Add synonyms where business language differs from the stored values. Connect related datasets when the answer spans more than one source. Then let users ask for the information they actually need.
They do not need to know the column names.
They do not need to know which dataset contains each field.
They do not need to build a formula for common summaries.
They ask the question.
Data Search finds the records and returns the answer.
